WO2006017700A2 - Analyte filter method and apparatus - Google Patents

Analyte filter method and apparatus Download PDF

Info

Publication number
WO2006017700A2
WO2006017700A2 PCT/US2005/027818 US2005027818W WO2006017700A2 WO 2006017700 A2 WO2006017700 A2 WO 2006017700A2 US 2005027818 W US2005027818 W US 2005027818W WO 2006017700 A2 WO2006017700 A2 WO 2006017700A2
Authority
WO
WIPO (PCT)
Prior art keywords
filter
analyte
frequency
spectrum
shape
Prior art date
Application number
PCT/US2005/027818
Other languages
French (fr)
Other versions
WO2006017700A3 (en
Inventor
Timothy L. Ruchti
Alexander D. Lorenz
Kevin H. Hazen
Original Assignee
Sensys Medical, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sensys Medical, Inc. filed Critical Sensys Medical, Inc.
Publication of WO2006017700A2 publication Critical patent/WO2006017700A2/en
Publication of WO2006017700A3 publication Critical patent/WO2006017700A3/en

Links

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/27Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands using photo-electric detection ; circuits for computing concentration
    • G01N21/274Calibration, base line adjustment, drift correction
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J3/00Spectrometry; Spectrophotometry; Monochromators; Measuring colours
    • G01J3/28Investigating the spectrum
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J3/00Spectrometry; Spectrophotometry; Monochromators; Measuring colours
    • G01J3/28Investigating the spectrum
    • G01J3/42Absorption spectrometry; Double beam spectrometry; Flicker spectrometry; Reflection spectrometry
    • G01J3/433Modulation spectrometry; Derivative spectrometry
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G01N21/31Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
    • G01N21/35Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
    • G01N21/359Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using near infrared light
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2201/00Features of devices classified in G01N21/00
    • G01N2201/12Circuits of general importance; Signal processing
    • G01N2201/129Using chemometrical methods
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2201/00Features of devices classified in G01N21/00
    • G01N2201/12Circuits of general importance; Signal processing
    • G01N2201/129Using chemometrical methods
    • G01N2201/1293Using chemometrical methods resolving multicomponent spectra

Definitions

  • the invention relates generally to biomedical methods and apparatus. More particularly, the invention relates to processing spectra to yield enhanced analyte property estimations. Still more particularly, the invention relates to generating and using an analyte filter designed to an analyte signal shape in the presence of interference, preferably in combination with multivariate analysis.
  • FIR farnesoid frequency response
  • HR infinite impulse response
  • Digital filtering techniques are used to reduce or remove background effects, to reduce or remove high frequency noise, and to enhance signal.
  • Specific sub-types of digital filtering have been previously described to process noninvasive near-infrared data to isolate chemical or physical information related to physiology, instrumentation, and/or the environment.
  • Specific types of digital filtering, such as derivatives and Fourier filters are discussed here.
  • Calibration transfer is a standardization procedure designed to eliminate a full recalibration and to maintain information residing in the existing model. Calibration transfer is useful because sources of variation in the instrument and environment are modeled in the development of a training or calibration set. Therefore, as the instrument or environment state changes the model components does not exactly match the current state.
  • One approach to calibration transfer is to generate a robust model that covers all future conditions.
  • Experimental design is used to develop a robust calibration.
  • parameters include measurement conditions, such as temperature and humidity, as well as analyte/constituent concentration distributions.
  • This approach is effective in controlled environments when the analyte signal-to-noise ratio is strong.
  • the technique is not efficient in terms of time and money.
  • the quality of the calibration is suspect in terms of inability to predict future conditions that need to be incorporated into the original calibration.
  • the technique does not readily allow incorporation of future conditions that are later identified without a new experimental design and development of a new or updated calibration.
  • Calibration transfer is used to compensate for changes to an axis, such as an x-axis or a y-axis.
  • an approach to x-axis stability is to provide, with each sample or on a daily basis, a spectrum of a standard that is used to determine the x-axis.
  • polystyrene is often used for adjustment of the x-axis in the near- infrared.
  • Additional near-infrared wavelength standards include rare earth oxides, such as holmium oxide, erbium oxide, and dysprosium oxide.
  • Each standard or reference provides multiple peaks that are used to set or adjust the x-axis, such as a wavelength axis. In its broadest sense, any material that yields known or reproducible peaks for a given state is usable as an x-axis standard.
  • Calibration transfer is also used to adjust or compensate for changes to a y-axis.
  • a y-axis is commonly adjusted with a reference standard.
  • diffuse reflectance standards in the near-infrared include polytetrafluoroethylene diffuse reflectance standards, such as diffuse reflectance standards that come with diffuse reflectances of 2, 5, 10, 20, 40, 60, 80, and 99%.
  • Diabetes is a chronic disease that results in improper production and use of insulin, a hormone that facilitates glucose uptake into cells. While a precise cause of diabetes is unknown, genetic factors, environmental factors, and obesity appear to play roles. Diabetics have increased risk in three broad categories: cardiovascular heart disease, retinopathy, and neuropathy. Complications of diabetes include: heart disease and stroke, high blood pressure, kidney disease, neuropathy (nerve disease and amputations), retinopathy, diabetic ketoacidosis, skin conditions, gum disease, impotence, and fetal complications. Diabetes is a leading cause of death and disability worldwide. Moreover, diabetes is merely one among a group of disorders of glucose metabolism that also includes impaired glucose tolerance and hyperinsulinemia, which is also referred to as hypoglycemia. It continues to be beneficial to have increased accuracy and precision of estimation of glucose concentration from noninvasive spectra.
  • the invention comprises a method and apparatus for preprocessing noninvasive spectra with an analyte filter that preferably rejects variations that are likely to be detrimental to the measurement system, while passing the signal that is most likely unique to the target analyte.
  • the analyte filtered data are used to estimate an analyte property, such as a glucose concentration, in the presence of noise, interferences, state changes, and/or across analyzers.
  • Figure 1 provides reference and associated estimated glucose concentration profiles according to the invention
  • Figure 2 illustrates a 27-point Savitsky-Golay first derivative filter
  • Figure 3 provides an analyte filter according to the invention
  • Figure 4 provides derivative processed glucose concentration estimations overlaid onto a Clarke error grid
  • Figure 5 provides combined filter processed glucose concentration estimations according to the invention overlaid onto a Clarke error grid
  • Figure 6 provides an analyte filter according to the invention
  • Figure 7 illustrates the relative filtering of different frequencies using a zero-order derivative
  • Figure 8 illustrates the relative filtering of different frequencies using a first order derivative
  • Figure 9 illustrates the relative filtering of different frequencies using a second order derivative
  • Figure 10 illustrates the relative filtering of different frequencies using a Gaussian filter
  • FIG 11 illustrates benefits of an analyte filter in terms of signal and noise according to the invention.
  • the noninvasive measurement of blood and tissue analytes is accomplished through the application of a multivariate calibration model to a noninvasive spectral measurement.
  • the developed model is designed to exploit the multivariate measurement and distinguish the analyte signal of interest from interfering signals, unaccounted for changes in the interference lead to degradation of the measurement performance.
  • Types of unmodeled interference that can increase measurement error include instrument noise and performance changes, measurement artifacts, interfering analytes, changes in the optical properties of the tissue, and inconsistent optical coupling to the tissue.
  • an analyte filter which rejects variations that are likely to be detrimental to the measurement system, while passing the signal that is most likely unique to the target analyte.
  • the filter is based upon shape information of the target analyte that is represented by the signal, net analyte signal, or extracted features preferably with compensation for interferences.
  • the filter is optionally implemented in several stages, each stage being designed and optimized according to at least one criterion.
  • the analyte filter is preferably further enhanced through a theoretical design that involves the use of interference models, in addition to shape information, to develop an optimized analyte filter.
  • the filter is optionally used with a photo-detector array (PDA) to enhance noise rejection efficiency as compared to systems that use a single element detector and to improve measurement performance.
  • PDA photo-detector array
  • an analyte filter is used that preferably passes analyte related signal while suppressing interferences manifested as both high and low frequency noise.
  • a data set is provided requiring multivariate analysis to derive an analyte property from a spectral data set.
  • the data set without modification, is processed separately first with a derivative filter and second with an analyte filter.
  • Enhanced analyte property estimation is demonstrated with the analyte filter approach compared to the derivative filter approach demonstrating the advantage of an analyte filter over a derivative filter.
  • the glucose analyzer included a sample module and a base module coupled by a communication bundle.
  • the sample module included a source, backreflector, and optics.
  • the communication bundle carried power and optical signal.
  • the base module included a grating and a linear photo diode array detector. Wavelength and intensity references were collected and used. In this case, the wavelength reference was polystyrene and the intensity reference was polytetrafluoroethylene.
  • the sample was a set of human forearms.
  • Calibration data were collected with both a fixed probe and a floating probe, bottom-up measurement sampling the volar aspect of forearms, where the probe had a single bundlet.
  • Prediction spectra were collected with a both a fixed probe and floating probe, in both bottom up and top down fiber probe configurations sampling both the volar and dorsal aspect of forearms with a single collection fiber. While the example is to a specific analyzer, the invention is applicable to data matrices generated from a wide number of related analyzers and sample sites, such as those described in U.S. patent application number 10/472,856, which is incorporated herein in its entirety by this reference thereto.
  • the analyzers used in this example were used to collect a calibration, monitoring, and independent estimation or prediction data set of noninvasive spectra with correlated glucose concentrations.
  • the calibration, monitoring, and independent estimation data sets are used with the derivative processing and analyte filtering approach below.
  • the calibration matrix represents 1109 spectra collected on a total of six subjects using two analyzers over an eight week period.
  • the monitoring data set includes 1549 spectra collected on six subjects using a total of two analyzers over a period of twenty weeks.
  • the estimation or prediction data matrix represents 188 samples from nine different subjects collected over a total of twelve visits using a total of seven analyzers over a period of multiple weeks.
  • Prediction data sets are preferably generated with analyzers not used to generate the calibration data set or multivariate model.
  • Multivariate models are typically generated or used with filtered or preprocessed spectra. Multivariate models extract information correlated with the reference values, such as glucose concentrations associated with each spectra. Multivariate models also correlate spectral features, such as baseline variation, noise, and interferences to the reference. Therefore, filters that pass features, such as a baseline deviation, correlated interference absorbance bands, and high-frequency noise, degrade the subsequent model performance in terms of precision, accuracy, and/or robustness. Therefore, an analyte filter designed to the frequency response of the analyte shape of the analyte and further designed to minimize interference benefits data analysis when multivariate models are employed.
  • the independent prediction data set collected with the instrumentation described, supra is represented.
  • the reference glucose concentrations are presented as a function of sample number, which is a nonlinear time axis.
  • the glucose concentration profiles cover a wide dynamic range and break correlations with parameters, such as time and temperature.
  • the independent reference concentrations are overlaid with glucose concentration estimations using: (1 ) an algorithm using an order derivative filter; and (2) an algorithm using an analyte filter. Both the derivative filter and analyte filter are described, infra.
  • the data set collected with the instrumentation described, supra, is processed with separate algorithms: one having an order derivative filter and another having an analyte filter. Both the order derivative filter and the analyte filter were generated using an optimization routine that used a calibration data set and a monitoring data set. The optimal order derivative filter and analyte filter were then each applied to an independent prediction data set. Results for both approaches are summarized herein.
  • the first order derivative approach preprocesses the spectra in three steps: 1 ) a 27- point first order derivative Savitsky-Golay convolution; 2) selection of a data matrix associated with the 1150 to 1850 nm spectral range; and 3) mean centering. Iterative preprocessing of the calibration and monitoring data sets while varying all of the derivative filter parameters results in the above described filter in combination with a principal component regression (PCR) using an optimized total of 44 factors.
  • PCR principal component regression
  • the resulting standard error of estimation which is also loosely referred to as a standard error of prediction (SEP), on the new samples is 30.0 mg/dL.
  • SEP standard error of prediction
  • the frequency response of the 27-point first order derivative filter is represented.
  • the filter has a peak response at approximately 0.025 normalized frequency units and has multiple sub-peaks that pass higher frequency signal and noise. As shown below, this filter is substantially different than the analyte filter as the order derivative filter leaks high frequency noise.
  • the independent prediction data set is analyzed.
  • the resulting glucose concentration estimations are presented in Figure 4, which is overlaid onto a Clarke error grid. A total of 74.5, 25.0, and 0.5% of the resulting glucose estimations fell into the A, B, and C regions of a Clarke error grid, respectively.
  • the resulting F-value is 3.75.
  • the analyte filter approach preprocesses the spectra in three steps: 1 ) processing with the analyte filter shown in Figure 3; 2) selection of a data matrix associated with the 1150 to 1850 nm spectral range; and 3) mean centering.
  • the analyte filter is represented in a frequency response format.
  • the analyte filter reduces baseline deviation and extensively attenuates high frequency noise.
  • the result is a filter that is substantially different than a zero, first, second, or higher order derivative filter.
  • This particular filter uses a 0.04 normalized frequency units low-pass filter in combination with a 0.02 normalized frequency units high-pass filter.
  • both the low-pass and high-pass filters are thirtieth order filters, though a large range of orders and cut-on or cut-off parameters are alternatively used.
  • the analyte filter is non-symmetrical and is represented as the product of the low-pass filter and the high-pass filter.
  • the independent prediction data set is analyzed.
  • the resulting standard error of estimation which is also loosely referred to as a standard error of prediction (SEP) 1 on the new samples is 23.4 mg/dL.
  • SEP standard error of prediction
  • the resulting glucose concentration estimations are presented in Figure 5, which is overlaid onto a Clarke error grid. A total of 81.9, 18.1 , and 0.0% of the resulting glucose estimations fell into the A, B, and C regions of a Clarke error grid, respectively.
  • the resulting F-value is 6.16.
  • the analyte filter outperforms the optimized derivative filter.
  • the improved success of the analyte filter approach is at least in part due to the removal of high frequency noise passed by the first-order derivative approach.
  • FIG. 6 a second example of an analyte filter is provided where low frequency noise is reduced, high frequency noise is greatly attenuated, and the filter is non-symmetrical.
  • the following section details how many non-analyte filters fail in one or more of these parameters.
  • FILTERS A subset of digital filtering spectra is processing spectra with an order-derivative, such as a zero, first, second, or higher order derivative. As demonstrated herein, derivative filters pass high frequency spectral information. Another subset of digital filtering spectra is processing spectra with a symmetrical filter, such as a Fourier filter. Both of these filter types use responses that are restricted or constrained. That is, setting a particular filter efficiency at two or more frequencies forces particular filter responses at other frequencies that is detrimental. This detrimental response is described in detail in the subsequent paragraphs.
  • Analyte filters do not carry the restraint of low degrees of freedom, of forcing a magnitude of a frequency response at an « th frequency based upon the magnitude of the frequency response at a first, second, third, or fourth frequency, or being forced to be symmetrical.
  • the three represented filters are all Savitsky-Golay zero-order derivative filters with 5, 15, and 35-point windows, respectively.
  • the 5-point zero-order derivative filter 601 is observed to have a high response at low frequency and to generally roll off at higher frequency.
  • the response of this filter is to pass low frequency information, such as baseline deviations, and to remove some high frequency information from the spectra, such as noise.
  • the 15-point zero-order derivative filter 602 again passes very low frequency information but has a much steeper roll-off. The steeper roll off results in a greater percentage removal of semi-high frequency noise.
  • the filter rings at higher frequency.
  • the 35-point zero-order derivative filter 603 again passes low frequency information and has a sharper cut-off. Generally, a zero-order derivative filter passes low frequency information, reduces high-frequency information, has a steeper negative slope with an increasing number of points, and leaks some high frequency response. All of these filters leak high-frequency noise and defining two points of a filter defines the magnitude at all frequencies for a given filter.
  • first-order derivative filters 700 presented in frequency space, are provided. The three represented filters are all Savitsky-Golay first-derivative filters with 5, 15, and 35-point windows, respectively.
  • the 5-point first-order derivative filter 701 strongly attenuates very low frequencies, is observed to have a peak response at a low-frequency, again attenuates to a degree at higher frequency, and has ripple at still greater frequency. The response of this filter is thus to attenuate very low frequency information greatly, such as baseline deviations, to pass low-frequency, and to again greatly attenuate high frequency noise with secondary ripple passing high frequency noise.
  • the 15-point first-order derivative filter 702 attenuates very low frequency information but has a narrower passband at lower frequency than the 5-point first-order derivative filter, and more ripples leaking high frequency noise.
  • the 35-point first-order derivative filter 703 attenuates to a degree low and high-frequency information, has a passband at relatively lower frequency, and has many ripples leaking higher frequency information, such as high-frequency noise.
  • a first-order derivative filter filters both high and low frequency information, has a passband that passes narrower regions at lower frequencies with an increase in the number of filter points in the convolution, leaks high frequency response through the ripples of the filter, and has a low degree of freedom.
  • second-order derivative filters 800 presented in frequency space.
  • the three represented filters are all Savitsky-Golay second-derivative filters with 5, 15, and 35-point windows, respectively.
  • the 5-point second-derivative filter 801 greatly attenuates low frequency, is observed to have a peak response at an intermediate frequency, again attenuates to a degree at higher frequency, and has ripple at still greater frequency.
  • the response of this filter is to attenuate low frequency information greatly, such as baseline deviations, to pass intermediate-frequency, and to attenuate high frequency noise somewhat with secondary ripple passing high frequency noise.
  • the 15-point second derivative filter 802 again greatly attenuates very low frequency information, but has a narrower passband at lower frequency than the 5-point second-derivative filter, and more ripples leaking high frequency noise.
  • the 35-point second derivative filter 803 attenuates low and high-frequency information, has a passband at relatively lower frequency, and has many ripples leaking higher frequency information, such as high-frequency noise.
  • a second-order derivative filter filters both high and low frequency information, has a passband that passes narrower regions at lower frequency with an increase in the number of filter points in the convolution, leaks high frequency response through the ripples of the filter, and has a low degree of freedom.
  • Higher order derivative filters have responses that are extended trends from the first and second derivative filters.
  • a higher order derivative filter filters both high and low frequency information, has a passband that passes narrower regions at lower frequency with an increase in the number of filter points in the convolution, leaks high frequency response through the ripples of the filter, and is restricted in terms of magnitude response across frequency.
  • Order-filters are restricted in terms of independent magnitude of response versus frequency. For example, if a desired filtering efficiency is set at two frequencies, then an order-filter that achieves this response necessarily has some filtering efficiency at another frequency. For example, forcing a filter to greatly attenuate very low frequencies results in ripples passing higher frequency. As a second example, forcing a narrower bandpass results in a greater number of ripples.
  • the restriction of one shape of the filter by defining another limits performance of the filters. Analyte filters are not limited by these restrictions.
  • a Gaussian filter reduces both baseline deviations occurring at low normalized frequency units, reduces high frequency noise occurring at high normalized frequency units, and has a passband over a range of intermediate normalized digital frequency units.
  • a Gaussian function is defined by its mean and standard deviation. This restricts the curvature of the of cut-on and the cut-off of the filter to be symmetrical. This symmetry restricts the flexibility of having a sharper cut-on versus the cut-off or vise versa. Referring now to Figure 10, an example of a Gaussian filter is provided having a mean of 0.2 and a standard deviation of 0.05.
  • Fourier filters operate by performing a Fourier transform of an absorbance spectrum as a function of wavelength to represent the data in intensity versus frequency.
  • the transformed spectrum multiplied by a Gaussian function and an inverse Fourier transform is subsequently performed to transform the spectrum back to its original domain.
  • this Gaussian filter type is constrained as a Gaussian filter is defined by a mean and a standard deviation so that the roll-off of one side of the filter necessarily impacts the roll-off of the opposite side of the filter.
  • a Gaussian filter is restricted. If two points are defined, then the rest of the Gaussian filter shape is defined.
  • the low degree of freedom restriction of shape and/or symmetry of a Gaussian filter is not present in an analyte filter. This allows the analyte filter to be designed or built based upon the actual frequency response of the analyte in the presence of noise and interference.
  • An analyte filter is a filter that preferably rejects variations that are likely to be detrimental to the measurement system, while passing the signal that is most likely unique to the target analyte.
  • an analyte filter based upon shape information of the target analyte which is based upon the analyte signal in the presence of noise and/or interference, is represented by the net analyte signal, or is based upon one or more extracted features.
  • a typical use of an analyte filter is to improve the robustness of a calibration with respect to sources of interference.
  • an analyte filter is built or optimized to a set of specifications that can include: the analyte signal, noise, and interference.
  • an analyte filter is used in combination with a multivariate soft model, such as partial least square or principal component regression, as soft models have difficulties compensating for noise, baseline variation, and interferences.
  • the analyte filter removes, minimizes, or reduces noise, baseline deviation, and interferences.
  • the analyte filter is optionally implemented in several stages, each stage designed and optimized according to a particular criterion.
  • an analyte filter is generated or enhanced through a theoretical design that involves the use of interference models, in addition to shape information, to develop an optimized analyte filter.
  • an iterative approach is used to generate or enhance an analyte filter shape, where the iterative approach uses knowledge of the analyte and/or noise frequency response of the system.
  • An analyte filter is not restricted in the frequency domain in the manner of a derivative leaking excess high frequency noise, is not restricted to a symmetrical shape, and has an n" point in the frequency domain not restricted by a combination of two, three, four, five or more points of the analyte filter in the frequency domain.
  • analyte filter Based upon the specification and/or design approach of an analyte filter, a number of approaches are used to generate an analyte filter for an application. Several example processes of analyte filter design are provided, infra.
  • An analyte filter is optionally created or optimized using the shape of the target analyte, preferably in conjunction with interference.
  • the pure component spectrum of a target analyte is often considered the signal of interest in applications involving spectroscopy and noninvasive analyte property determination.
  • the use of preprocessing techniques is traditionally aimed at enhancing and extracting the pure component analyte signal.
  • the true signal of interest in multivariate applications is contained in the shape of the target analyte after orthogonalization with respect to the interference.
  • the net analyte signal defines that portion of the pure component analyte spectrum that is accessible for analyte measurement and is often significantly different than the pure component spectrum related to the target analyte.
  • shape is the signal in the presence of interference.
  • One measure of signal strength is the net analyte signal, nas, given by the projection of the pure component analyte signal, R, onto the null space of the interference contained in the B, where / is the identity matrix as provided in equation 1.
  • the interference, B is a basis set that spans at least a portion of the space of expected variation.
  • nas is estimated from the regression vector of the multivariate calibration.
  • the net analyte signal constitutes a pattern, contained in the vector nas, with shape information that is used to determine the optimal filter characteristics.
  • the frequency response of the nas denoted nas ⁇ , provides the magnitude and phase of the signal of interest versus the normalized frequency and is used to establish filter criteria yielding minimal attenuation of the signal of interest.
  • a filter is designed to pass frequency components of the nas that are significant while rejecting harmonics that contribute minimally to the analyte property estimation.
  • a primary interference affecting the analyte measurement is related to low frequency harmonics that develop due to instrument drift, instrument-to- instrument differences, and/or gross changes in the scattering properties of the tissue between subjects and over time.
  • the enhancement of analytical signal, optionally measured by the n ⁇ s, is performed by rejecting low frequency interference and passing frequency components that has the highest probability of representing the signal of interest in subsequent samples.
  • a high-pass filter is implemented with a break frequency or cut-on, ⁇ ., that is established on the basis of n ⁇ s ⁇ .
  • the selection of break frequency is based upon the point at which ⁇ n ⁇ s(f) ⁇ falls below a minimum threshold level, which is typically about an order of magnitude less than the maximum frequency component though of fractional order of magnitude or two or more orders of magnitude less thresholds is optionally used.
  • a minimum threshold level typically about an order of magnitude less than the maximum frequency component though of fractional order of magnitude or two or more orders of magnitude less thresholds is optionally used.
  • specific features of the analyte signal, the signal in the presence of interference, or the nas are extracted to provide further information to enhance the design of the analyte filter. For example, the most influential regions of the nas are characterized according to wavelength range, frequency content, and bandwidth. Optionally, these features are extracted using preprocessing techniques, described infra. Optimal filters are employed as a function of the influential individual regions yielding a
  • a-priori knowledge of the measured system is used in the creation and/or optimization of an analyte filter.
  • the use of a-priori knowledge for the development of an analyte filter is accomplished by selecting, parameterizing, and implementing an optimized filter using what is known about noise, interference, and the signal of interest. This method does not directly rely on a set of empirical data for the determination of the analyte filter, but instead uses theoretical methods to make a robust and optimal system for separating the spectroscopic signal from noise and interference.
  • a-priori knowledge is used in conjunction with, subsequent to, concurrently with, or iteratively with empirical data in analyte filter creation.
  • the a-prior information about random noise is in the form of one or more models, such as a statistical model, that are used to provide the corresponding frequency characteristics, N(f) .
  • the a-priori information about interfering analytes is provided through an estimate of the net analyte signal, extracted features, interferences, and/or noise characteristics, which provide shape information about the signal of interest.
  • A-priori knowledge in analyte filter design is optionally used in combination with preprocessing.
  • the low frequency variation over the wavelength axis of the measured spectrum is attenuated through a high-pass filtering operation.
  • the cut-off frequency or bandwidth of the high-pass operation is determined according to the shape information of the nas as previously discussed.
  • the second function is a low-pass or smoothing operation in which noise is suppressed through the attenuation of high frequencies.
  • the break frequency of the low-pass filter is set according to the bandwidth of the spectrometer, the net analyte signal of the target analyte or constituent, and/or the necessary signal-to- noise ratio required to make the measurement. For example, a-priori knowledge of the system is used to generate a two-stage filter.
  • the measured spectrum is oversampled with respect to the wavelength axis and the low-pass bandwidth is set based upon an analysis of the signal-to-noise ratio where the net analyte signal is the signal and the noise is the root-mean-square variation of the measured spectrum in the wavelength region used for measurement of the target analyte.
  • the low-pass bandwidth is reduced, the high frequency components of the noise are attenuated, leading to a reduction in the root-mean-square noise.
  • this process also attenuates the high frequency components of the signal leading to a simultaneous reduction in the net analyte signal.
  • the noise is distributed in greater proportions at higher frequencies than the net analyte signal. Therefore, low-pass filtering the measured spectrum removes a greater proportion of the noise than the net analyte signal.
  • the optimal low-pass bandwidth can be defined as one that maximizes the ratio of the signal or net analyte signal to the noise. Given a frequency domain model the net analyte signal, NAS(f) , and the noise, N(f) , the bandwidth of the low-pass filter, f bw , can be determined through equation 2
  • the break frequency of the high-pass section is set to attenuate low frequency variation caused by changes in the scattering while passing the net analyte signal. This is generally accomplished empirically through an exemplary set of data or through a harmonic analysis of the net analyte signal.
  • signal as opposed to net analyte signal is used in this optimization.
  • the analyte filter is employed to compensate for instrument related shape distortion that causes a contraction, dilation, or shift.
  • a-prior knowledge of wavelengths of absorbance of a particular feature are optionally used as standards for a wavelength correction.
  • a Bessel-Thomson filter is used to provide a constant correction with respect to wavelength. The measurement of a complete spectrum enables that filter to be either causal or non-causal, which provide the opportunity to shift a spectrum in either direction along an axis.
  • complex structures or adaptive means are employed for adjusting the filter characteristics on the basis of an acquired signal representative of the distortion. Adaptive means for filtering using wavelets, artificial neural networks, a Kalman filter, and related methods are also used, depending on the application.
  • an analyte filter is used to take advantage of known hardware features.
  • many existing spectrometer systems use a single element detector as a means for generating a response, such as a near- infrared spectrum.
  • a principle problem with this approach is that the serial extraction of intensity over time leads to a wavelength-to-wavelength correlation in the noise. Consequently, the frequency response of the measurement noise is dominated by low-frequency harmonics that can not be efficiently filtered without also attenuating the signal of interest.
  • an array-based system is employed, where a multiplicity of detectors convert spatially distributed electro-magnetic energy into a spectrum suitable for processing. Each element detects a band of wavelengths and operates independently.
  • an analyte filter optimized for the same matrix with the same chemical and environment interferences is different for an array based spectrometer and a single element detector based spectrometer.
  • Most analytes in visible, near-infrared, or infrared spectroscopy are dominated by low frequencies.
  • a single element detector is characterized by an increasing magnitude of noise as the normalized frequency decreases 1 101.
  • the frequency response of an array detector is flat or nearly flat 1102. This means that the lower frequency noise of the array detector is de-emphasized.
  • an analyte filter optimized for a glucose signal using an array detector optimally shifts to lower frequencies 1103, compared to an analyte filter optimized for a glucose signal using a single element detector 1104.
  • molecular analytes have spectral signatures dominated by lower frequencies.
  • a near-infrared glucose signal in frequency space 1105 exemplifies this.
  • the analyte filter optimized using the flat noise characteristic moves closer to the low frequencies that dominate the glucose signal.
  • the analyte filter generated using an array detector is different and more efficient compared to the analyte filter optimized for the glucose signal using a single element detector.
  • the lower frequency analyte filter optimized with a detector array noise structure allows extraction of a greater portion of the glucose signal, resulting in improved performance of the analyte concentration estimation using noninvasive spectra.
  • analyte filter associated with a particular cluster of samples that represent a more uniform tissue state.
  • the analyte filter is preferably associated with a particular calibration that is specific to the variation represented by the cluster.
  • an adaptive filter that, in addition to the more general analyte filter, removes or attenuates spectral interference that is present in a particular cluster.
  • the filter operates by using a basis set of background spectra, such as interference spectra, to reduce the contribution of the cluster specific interference to future measurements.
  • a data set is used for calibration including x c G ⁇ PXN and y c e $RTM , where x c is the matrix of tissue measurements, y c is a vector of analyte concentrations, P is sample number, and N is data point, such as wavelength, where ⁇ c and y c are collected or associated with a particular state.
  • a state is defined by one or more of the following: subject, day, time, tissue position, temperature, environmental condition, tissue properties, instrument state, etc.
  • the adaptive filter uses a set of background spectra, which define the interference to optimally reduce subsequent contributions of similarly shaped interference.
  • A I where y c is the reference concentration, the k lh element of y c ⁇ s, y c k , and L is the set of spectra used to calculated x b .
  • a basis set is determined using the first one or more spectra associated with a particular condition. Subsequent samples are adaptively filtered using the method above until a change in the interference is detected.
  • the analyte filter is used to adapt a spectral reading to a nearest cluster, to translate a reading from one cluster to another cluster, and to determine outliers as described in U.S. patent application number 09/664,973 filed September 18, 2000 and in U.S. patent application number 10/870,727 filed June 16, 2004 both of which are incorporated herein in their entirety by this reference thereto.
  • Example Vl Multi-Stage Filter An analyte filter is optionally designed to the signal of interest compensating for noise structure without restraint on the analyte filter shape.
  • Existing filter methodologies for noninvasive detection of analytes are limited as a consequence of the underlying filter structure.
  • the design of the optimal analyte filter involves the use of multiple design goals, existing filter implementations constrain the high-pass and low-pass filter characteristics according to a deterministic function. For example, filtering involves the use of functions, such as the Savitsky-Golay derivative, which contain significant stop-band ripple and thereby pass unwanted noise. Passing unwanted high frequency noise results in performance degradation, as described supra.
  • a finite-impulse response filter is designed and implemented by constraining the frequency response according to a Gaussian curve. While the resulting step response with respect to the wavelength axis is well behaved, the low and high-pass filter characteristics are necessarily sub-optimal due to symmetry restrictions.
  • other methods of filtering involve the use of functions, such as the Savitsky-Golay derivative, which contain significant stop-band ripple and thereby pass unwanted noise, which leads to performance degradations.
  • the problems of a restrained filter or a filter leaking excess high frequency noise are addressed.
  • the generation of a filter that is optimized according to the shape information contained in the signal of the analyte, such as the net analyte signal, and the noise is preferably performed with at least two design goals and involves a stage for each goal.
  • optimization of the signal is preferably performed with optimization of a longpass filter or with a highpass filter designed to pass the signal or net analyte signal frequency response shape.
  • a filter is optimized for the noise structure.
  • a shortpass or lowpass filter is used or optimized based upon the noise structure so that high frequency noise is reduced, minimized, or eliminated.
  • the signal and noise optimization functions are optionally performed in either order depending on the implementation of the band-pass filter.
  • the methods used for performing this operation include infinite-impulse response (HR) and finite- impulse response (FIR) band-pass filtering.
  • a FIR filter is implemented according to equation 5,
  • m f J is the filtered spectrum at the / h measured wavelength
  • m y is the measured spectrum at the/ h wavelength
  • a k denotes the k th filter coefficient
  • P is the length of the filter impulse response or filter window width.
  • the filter is non-causal and applied across the wavelength axis of the measured spectrum.
  • the filter width, P is assumed to be an odd number and the filter coefficients are determined according to the desired filter break frequencies and characteristics of the pass and stop-bands. This multiple stage filter approach reduces and in some instances eliminates the restraints of traditional filters.
  • Analyte filters are optionally built based upon specifications and/or shape. For example, combinations of longpass, shortpass and Butterworth filters are used to meet objectives set out in the specification requirements of a given design filter. Combinations of filters are created resulting in a complex filter, a two-stage filter, a three stage-filter, or an ⁇ th stage filter.
  • non-symmetrical filters such as shortpass filters, longpass filters, bandpass or passband filters, and MR Butterworth filters.
  • these filter types are not customarily used in combination to create an analyte filter shape based upon a set of specifications. Combinations of these filters minimize high frequency leaking, reduce baseline deviation, avoid symmetry concerns, and match, correlate, or correspond to the specified shape of the analyte frequency response in the presence of interference.
  • the order of the filters is varied.
  • a complex filter has a simple or complex shape.
  • complex filters are not restricted to the limitations of an order derivative filter or a Gaussian filter.
  • the filters need not be symmetrical and do not necessarily pass higher frequency intensity noise.
  • a complex filter is generated in many ways.
  • a complex filter is optimized using a response function, theoretically generated using net analyte signal, designed for noise reduction, or is a complex shape designed from multiple inputs.
  • an analyte filter is generated using data from a first set of analyzers or spectrometers.
  • An additional analyzer or a second set of analyzers that are not present in the first set of analyzers have distinct spectral characteristics that can degrade analytical performance of analyte property determination.
  • Implementation of a model built on the first set of analyzers onto the second set of analyzers thus requires a calibration transfer or filter for optimal performance.
  • Calibration transfer is a standardization procedure designed to eliminate a full recalibration and to maintain information residing in the existing model. Calibration transfer is useful because in the development of the training set, also referred to as a calibration set, sources of variation in the instrument, sample, and environment are modeled. Therefore, as the instrument, sample, or environment state changes the model components do not exactly match the current state. These problems are typically compounded when transferring a calibration from a first spectrometer to a second spectrometer. Calibration transfer addresses a range of concerns including: wavelength or x-axis stability, energy throughput or y-axis stability, and bandwidth. Calibration transfers are used across time on one or more instruments, across many instruments of the same design, and across instruments of different design.
  • Calibration transfer is also used when the sample varies.
  • a living organism such as the human body, undergoes continuous change.
  • calibration transfers are used to adjust for changes in the environmental conditions, such as changes in humidity or temperature. While broad models account for many changes, changing state still leads to an increase in estimation error or analyte determination.
  • spectrometers There are many potential sources of variation in an optical based analyzers, which are referred to herein as spectrometers or analyzers.
  • a first source of variation is when there is a change in the entire spectrometer design. This is the case, for example, when a calibration is built on a first spectrometer, such as a master spectrometer and estimations or predictions are performed on a different spectrometer, such as a slave spectrometer. This first type of variation is common when a research grade spectrometer is used for calibration and a process grade spectrometer, built with less stringent specifications, is used for subsequent analyses.
  • a second source of variation is a change in a part of a spectrometer.
  • changing a part of a spectrometer include: changing a source, which can effect the blackbody radiation emitted; changing a monochromator or grating, which can effect the wavelength axis; changing a fiber optic or fiber bundle, which can effect the bandwidth; or changing the detector, which can effect the response curve or change a high or low frequency cut-off.
  • a third source of variation results from alignment of the spectrometer. Alignment changes result from movement of one or more of sub-components of the spectrometer. Alignment issues also result from mechanical design and fabrication tolerances.
  • a fourth source of variation is change in the environmental conditions about the spectrometer or analyzer.
  • Common environmental changes affecting near-infrared spectra include temperature and humidity.
  • a fifth source of variation is a change in the sample.
  • changes in the sample include changes in chemical composition, sample finish, particle size distribution, and density.
  • changes in a skin sample include changes in temperature, chemical composition, chemical distribution, physical structure, hydration, and/or localized pressure.
  • the state of the spectrometer affects its output. Variation in state results in variation of output. For example, variation in state results in variation of the observed or calculated absorption coefficient and/or the observed or calculated reduced scattering coefficient. Calibration transfer is useful for adjusting for or compensating for the change in output resulting from change in state.
  • an analyte filter is used advantageously in the calibration transfer process.
  • the analyte filter is used to extract the signal from a known type of interference or from a particular cluster of sample as provided herein.
  • the analyte filter is optionally used with a preprocessing and/or processing step. Several preprocessing techniques are described, infra.
  • data preprocessing typically the same preprocessing is performed on data that are used to generate a calibration model and the data that the model applied to.
  • Preprocessing techniques include: multiplicative scatter correction, standard normal variate transformation, finite impulse response, a Bessel filter, normalization, a moving average, a finite impulse response filter, an infinite impulse response filter, and Fourier transformation.
  • An additional preprocessing technique is extraction of an absorbance and/or scattering signal with one or more sub-ranges of a spectrum. The selected and optionally further modified signal is preferably subsequently filtered with an analyte filter.

Abstract

The invention comprises a method and apparatus for enhancing the analysis of noninvasive spectra, resulting in improved analytical performance. More particularly, the invention comprises a method and apparatus for processing noninvasive spectra with an analyte filter that preferably rejects variation likely to be detrimental to the measurement system, while passing signal that probabilistically is unique to the target analyte. Subsequently, the analyte filtered data are used to estimate an analyte property, such as a glucose concentration, in the presence of noise, interferences, state changes, and/or across analyzers.

Description

ANALYTE FILTER METHOD AND APPARATUS
BACKGROUND OF THE INVENTION TECHNICAL FIELD
The invention relates generally to biomedical methods and apparatus. More particularly, the invention relates to processing spectra to yield enhanced analyte property estimations. Still more particularly, the invention relates to generating and using an analyte filter designed to an analyte signal shape in the presence of interference, preferably in combination with multivariate analysis.
DESCRIPTION OF RELATED ART
DIGITAL FILTERING
The technique of digitally filtering collected data is performed to isolate a portion of the data. Digital filtering is broadly broken into two types, finite impulse response
(FIR) and infinite impulse response (HR). Digital filtering techniques are used to reduce or remove background effects, to reduce or remove high frequency noise, and to enhance signal. Specific sub-types of digital filtering have been previously described to process noninvasive near-infrared data to isolate chemical or physical information related to physiology, instrumentation, and/or the environment. Specific types of digital filtering, such as derivatives and Fourier filters are discussed here.
Derivative / Convolution
A. Savitsky and M. Golay, Smoothing and differentiation of data by simplified least squares procedures, Anal. Chem., 36, 1627-40, (1964) describe a convolution based implementation of smoothing, first derivative filtering, second derivative filtering, and higher order derivative filtering for enhancing the signal-to-noise ratio of a response or a vector. D. Haaland, M. Robinson, G. Koepp, E. Thomas, and R. Eaton, Reagentless near- infrared determination of glucose in whole blood using multivariate calibration, Appl. Spect, 46, 1575-1578, (1992) describe the use of derivatives in conjunction with glucose concentration determination. The authors suggest use of derivative spectra for the reduction of subject-to-subject or inter-subject spectral variation.
J. Hall, Method and device for measuring concentration levels of blood constituents non-invasively, U.S. patent no. 5,361 ,758 (November 8, 1994) describes the use of a derivative in processing of near-infrared noninvasive spectra. J. Samsoondar, Method for calibrating spectrophotometric apparatus with synthetic fluids to measure plasma and serum analytes, U.S. patent no. 6,470,279 (October 22, 2002); J. Samsoondar, Method for calibrating spectrophotometric apparatus, U.S. patent no. 6,611 ,777 (August 6, 2003); and J. Samsoondar, Method for calibrating spectrophotometric apparatus, U.S. patent no. 6,651 ,015; November 18, 2003) describe the use of order derivatives to process spectra. The processed spectra are used on spectra collected on a second instrument to enhance calibration transfer developed on at least one additional instrument. In this document, an order derivative is also referred to as a mathematical derivative.
Fourier Filtering
M. Arnold, et. al., Determination of physiological levels of glucose in an aqueous matrix with digitally filtered Fourier transformed near-infrared spectra, Anal. Chem., 62, 1457-1464, (1990); G. Small, et. al. Strategies for Coupling Digital Filtering with partial least-squares regression: application to the determination of glucose in plasma by Fourier transform near-infrared spectroscopy, Anal. Chem, 65, 3279- 3289, (1993); G. Small, M. Arnold Method and apparatus for non-invasive detection of physiological chemicals, particularly glucose, U.S. patent no. 5,459,317 (October 17, 1995); and G. Small, M. Arnold, Method and apparatus for non-invasive detection of physiological chemicals, particularly glucose, U.S. patent no. 6,061 ,582 (May 9, 2000) describe the use of a Gaussian digital filter for processing absorbance data after transforming from a wavelength to a frequency domain. CALIBRATION TRANSFER Overview
Calibration transfer is a standardization procedure designed to eliminate a full recalibration and to maintain information residing in the existing model. Calibration transfer is useful because sources of variation in the instrument and environment are modeled in the development of a training or calibration set. Therefore, as the instrument or environment state changes the model components does not exactly match the current state.
Identical performance of analytical instruments is unrealistic even with the successful implementation of tight quality control on instrument hardware. For example, variation in the output of a source, quality of lenses or mirrors, alignment, and detector response, which are limited by manufacturing tolerances, result in differences between spectrometers even of the same design. The instrument differences result in spectra of the master instrument varying from that of the slave instrument. Variations between the spectrometers result in errors when using a calibration developed on a spectrometer to determine parameters with a second spectrometer. Generally, this error is increasingly detrimental as the signal-to-noise ratio of the determined analyte decreases. Several techniques for calibration transfer are presented here.
Robustness
One approach to calibration transfer is to generate a robust model that covers all future conditions. Experimental design is used to develop a robust calibration. For noninvasive glucose concentration determinations, parameters include measurement conditions, such as temperature and humidity, as well as analyte/constituent concentration distributions. This approach is effective in controlled environments when the analyte signal-to-noise ratio is strong. However, the technique is not efficient in terms of time and money. Also, the quality of the calibration is suspect in terms of inability to predict future conditions that need to be incorporated into the original calibration. In addition, the technique does not readily allow incorporation of future conditions that are later identified without a new experimental design and development of a new or updated calibration.
Full Recalibration Full recalibration of an analyzer is not preferable due to time requirements, technical expertise requirements, and expense. In addition, recalibration often fails to capture a full range of parameters, such as variations in the environment and instrument, thereby forcing additional recalibrations as the state of these parameters change.
Axis Standardization
Calibration transfer is used to compensate for changes to an axis, such as an x-axis or a y-axis. For spectrophotometric based determination, an approach to x-axis stability is to provide, with each sample or on a daily basis, a spectrum of a standard that is used to determine the x-axis. For adjustment of the x-axis in the near- infrared, polystyrene is often used. Additional near-infrared wavelength standards include rare earth oxides, such as holmium oxide, erbium oxide, and dysprosium oxide. Each standard or reference provides multiple peaks that are used to set or adjust the x-axis, such as a wavelength axis. In its broadest sense, any material that yields known or reproducible peaks for a given state is usable as an x-axis standard.
Calibration transfer is also used to adjust or compensate for changes to a y-axis. For example, a y-axis is commonly adjusted with a reference standard. Examples of diffuse reflectance standards in the near-infrared include polytetrafluoroethylene diffuse reflectance standards, such as diffuse reflectance standards that come with diffuse reflectances of 2, 5, 10, 20, 40, 60, 80, and 99%.
Another approach is the use of standards that simulate the target sample, such as a tissue phantom or intralipid. In its broadest sense, any material that yields known or reproducible transmittance, reflectance, or diffuse reflectance is usable as a y-axis standard. A number of difficulties exist for remeasuring standards. First, instability of the sample creates difficulties in producing a spectrum that is constant across time. Second, reproducing the environment, which affects the resulting spectra is difficult. For example, temperature and humidity effect spectra. Third, movement of the analyzer is an issue due to alignment. For example, this is relevant when the analyzer is moved from a lab to a process line or from a laboratory or production facility to a hospital or home setting. Replacement of analyzer components also leads to generation of spectra that are not reproducible.
DIABETES
Diabetes is a chronic disease that results in improper production and use of insulin, a hormone that facilitates glucose uptake into cells. While a precise cause of diabetes is unknown, genetic factors, environmental factors, and obesity appear to play roles. Diabetics have increased risk in three broad categories: cardiovascular heart disease, retinopathy, and neuropathy. Complications of diabetes include: heart disease and stroke, high blood pressure, kidney disease, neuropathy (nerve disease and amputations), retinopathy, diabetic ketoacidosis, skin conditions, gum disease, impotence, and fetal complications. Diabetes is a leading cause of death and disability worldwide. Moreover, diabetes is merely one among a group of disorders of glucose metabolism that also includes impaired glucose tolerance and hyperinsulinemia, which is also referred to as hypoglycemia. It continues to be beneficial to have increased accuracy and precision of estimation of glucose concentration from noninvasive spectra.
There remains an unsolved need for extracting data from spectral data, such as noninvasive spectra, that is useful in generating subsequent analyte property estimations. It would be advantageous to provide a method and apparatus for enhancing the analysis of noninvasive spectra, resulting in improved analytical performance within an instrument, across instruments, and across states using an analytical filter. SUMMARY OF THE INVENTION
A method and apparatus for enhancing the analysis of noninvasive spectra resulting in improved analytical performance is presented. More particularly, the invention comprises a method and apparatus for preprocessing noninvasive spectra with an analyte filter that preferably rejects variations that are likely to be detrimental to the measurement system, while passing the signal that is most likely unique to the target analyte. Subsequently, the analyte filtered data are used to estimate an analyte property, such as a glucose concentration, in the presence of noise, interferences, state changes, and/or across analyzers.
BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 provides reference and associated estimated glucose concentration profiles according to the invention;
Figure 2 illustrates a 27-point Savitsky-Golay first derivative filter;
Figure 3 provides an analyte filter according to the invention;
Figure 4 provides derivative processed glucose concentration estimations overlaid onto a Clarke error grid;
Figure 5 provides combined filter processed glucose concentration estimations according to the invention overlaid onto a Clarke error grid;
Figure 6 provides an analyte filter according to the invention; Figure 7 illustrates the relative filtering of different frequencies using a zero-order derivative;
Figure 8 illustrates the relative filtering of different frequencies using a first order derivative;
Figure 9 illustrates the relative filtering of different frequencies using a second order derivative;
Figure 10 illustrates the relative filtering of different frequencies using a Gaussian filter; and
Figure 11 illustrates benefits of an analyte filter in terms of signal and noise according to the invention.
DETAILED DESCRIPTION OF THE INVENTION
The noninvasive measurement of blood and tissue analytes is accomplished through the application of a multivariate calibration model to a noninvasive spectral measurement. Although the developed model is designed to exploit the multivariate measurement and distinguish the analyte signal of interest from interfering signals, unaccounted for changes in the interference lead to degradation of the measurement performance. Types of unmodeled interference that can increase measurement error include instrument noise and performance changes, measurement artifacts, interfering analytes, changes in the optical properties of the tissue, and inconsistent optical coupling to the tissue.
To improve the robustness of a calibration with respect to sources of interference an analyte filter has been developed, which rejects variations that are likely to be detrimental to the measurement system, while passing the signal that is most likely unique to the target analyte. The filter is based upon shape information of the target analyte that is represented by the signal, net analyte signal, or extracted features preferably with compensation for interferences. To enhance the analyte filter functionality, the filter is optionally implemented in several stages, each stage being designed and optimized according to at least one criterion. The analyte filter is preferably further enhanced through a theoretical design that involves the use of interference models, in addition to shape information, to develop an optimized analyte filter. Finally, the filter is optionally used with a photo-detector array (PDA) to enhance noise rejection efficiency as compared to systems that use a single element detector and to improve measurement performance.
In a preferred embodiment of the invention, an analyte filter is used that preferably passes analyte related signal while suppressing interferences manifested as both high and low frequency noise. To illustrate the advantage of an analyte filter over a classical derivative type filter and to demonstrate how an analyte filter is used, an example is provided where noninvasive glucose concentration predictions are made using an algorithm. The example demonstrates that substitution of an analyte filter for a derivative processing step results in improved analytical performance. Further discussion follows Example I, which illustrates similar limitations in other filters. Still further discussion following Example I describes an analyte filter in more depth. The use of near-infrared noninvasive glucose concentration estimation in the example is illustrative in nature and is not intended to limit the scope of the invention.
Example I
A data set is provided requiring multivariate analysis to derive an analyte property from a spectral data set. The data set, without modification, is processed separately first with a derivative filter and second with an analyte filter. Enhanced analyte property estimation is demonstrated with the analyte filter approach compared to the derivative filter approach demonstrating the advantage of an analyte filter over a derivative filter.
Instrumentation A diffuse reflectance based glucose analyzer was used to collect calibration and estimation, also referred to as prediction, near-infrared spectra. The glucose analyzer included a sample module and a base module coupled by a communication bundle. The sample module included a source, backreflector, and optics. The communication bundle carried power and optical signal. The base module included a grating and a linear photo diode array detector. Wavelength and intensity references were collected and used. In this case, the wavelength reference was polystyrene and the intensity reference was polytetrafluoroethylene. The sample was a set of human forearms. Calibration data were collected with both a fixed probe and a floating probe, bottom-up measurement sampling the volar aspect of forearms, where the probe had a single bundlet. Prediction spectra were collected with a both a fixed probe and floating probe, in both bottom up and top down fiber probe configurations sampling both the volar and dorsal aspect of forearms with a single collection fiber. While the example is to a specific analyzer, the invention is applicable to data matrices generated from a wide number of related analyzers and sample sites, such as those described in U.S. patent application number 10/472,856, which is incorporated herein in its entirety by this reference thereto.
Data Set The analyzers used in this example were used to collect a calibration, monitoring, and independent estimation or prediction data set of noninvasive spectra with correlated glucose concentrations. The calibration, monitoring, and independent estimation data sets are used with the derivative processing and analyte filtering approach below. The calibration matrix represents 1109 spectra collected on a total of six subjects using two analyzers over an eight week period. The monitoring data set includes 1549 spectra collected on six subjects using a total of two analyzers over a period of twenty weeks. The estimation or prediction data matrix represents 188 samples from nine different subjects collected over a total of twelve visits using a total of seven analyzers over a period of multiple weeks. Prediction data sets are preferably generated with analyzers not used to generate the calibration data set or multivariate model. Data Analysis
Multivariate models are typically generated or used with filtered or preprocessed spectra. Multivariate models extract information correlated with the reference values, such as glucose concentrations associated with each spectra. Multivariate models also correlate spectral features, such as baseline variation, noise, and interferences to the reference. Therefore, filters that pass features, such as a baseline deviation, correlated interference absorbance bands, and high-frequency noise, degrade the subsequent model performance in terms of precision, accuracy, and/or robustness. Therefore, an analyte filter designed to the frequency response of the analyte shape of the analyte and further designed to minimize interference benefits data analysis when multivariate models are employed.
Referring now to Figure 1 , the independent prediction data set collected with the instrumentation described, supra, is represented. The reference glucose concentrations are presented as a function of sample number, which is a nonlinear time axis. The glucose concentration profiles cover a wide dynamic range and break correlations with parameters, such as time and temperature. The independent reference concentrations are overlaid with glucose concentration estimations using: (1 ) an algorithm using an order derivative filter; and (2) an algorithm using an analyte filter. Both the derivative filter and analyte filter are described, infra.
The data set collected with the instrumentation described, supra, is processed with separate algorithms: one having an order derivative filter and another having an analyte filter. Both the order derivative filter and the analyte filter were generated using an optimization routine that used a calibration data set and a monitoring data set. The optimal order derivative filter and analyte filter were then each applied to an independent prediction data set. Results for both approaches are summarized herein.
The first order derivative approach preprocesses the spectra in three steps: 1 ) a 27- point first order derivative Savitsky-Golay convolution; 2) selection of a data matrix associated with the 1150 to 1850 nm spectral range; and 3) mean centering. Iterative preprocessing of the calibration and monitoring data sets while varying all of the derivative filter parameters results in the above described filter in combination with a principal component regression (PCR) using an optimized total of 44 factors. The resulting standard error of estimation, which is also loosely referred to as a standard error of prediction (SEP), on the new samples is 30.0 mg/dL. Within this data analysis, a 27-point first order derivative filter is determined to be optimal.
Referring now to Figure 2, the frequency response of the 27-point first order derivative filter is represented. The filter has a peak response at approximately 0.025 normalized frequency units and has multiple sub-peaks that pass higher frequency signal and noise. As shown below, this filter is substantially different than the analyte filter as the order derivative filter leaks high frequency noise. Using this optimal first order derivative filter and associated processing, the independent prediction data set is analyzed. The resulting glucose concentration estimations are presented in Figure 4, which is overlaid onto a Clarke error grid. A total of 74.5, 25.0, and 0.5% of the resulting glucose estimations fell into the A, B, and C regions of a Clarke error grid, respectively. The resulting F-value is 3.75.
The analyte filter approach preprocesses the spectra in three steps: 1 ) processing with the analyte filter shown in Figure 3; 2) selection of a data matrix associated with the 1150 to 1850 nm spectral range; and 3) mean centering. Referring now to Figure 3, the analyte filter is represented in a frequency response format. The analyte filter reduces baseline deviation and extensively attenuates high frequency noise. The result is a filter that is substantially different than a zero, first, second, or higher order derivative filter. This particular filter uses a 0.04 normalized frequency units low-pass filter in combination with a 0.02 normalized frequency units high-pass filter. In this example, both the low-pass and high-pass filters are thirtieth order filters, though a large range of orders and cut-on or cut-off parameters are alternatively used. The analyte filter is non-symmetrical and is represented as the product of the low-pass filter and the high-pass filter. As above, optimizing the filter parameters while optimizing the multivariate model using a calibration and monitoring data set results in a principal component regression model using an optimized total of 36 factors.
Using this analyte filter and the associated processing, the independent prediction data set is analyzed. The resulting standard error of estimation, which is also loosely referred to as a standard error of prediction (SEP)1 on the new samples is 23.4 mg/dL. The resulting glucose concentration estimations are presented in Figure 5, which is overlaid onto a Clarke error grid. A total of 81.9, 18.1 , and 0.0% of the resulting glucose estimations fell into the A, B, and C regions of a Clarke error grid, respectively. The resulting F-value is 6.16. Comparing the results from the analyte filter approach and the first-order derivative approach results in glucose estimations with 22% less error, a higher F-value, a greater percentage of data points resulting in the desirable A region of the Clarke error grid and no data points falling into the undesirable C region of the Clarke error grid. Clearly, the analyte filter outperforms the optimized derivative filter. The improved success of the analyte filter approach is at least in part due to the removal of high frequency noise passed by the first-order derivative approach.
Of particular importance is the fact that a variety of analyte filters are produced that have the common characteristics of greatly attenuating high normalized frequencies, reducing low normalized frequencies, having a peak response near the analyte frequency response, and not being forced into a symmetrical or constrained shape.
Referring now to Figure 6, a second example of an analyte filter is provided where low frequency noise is reduced, high frequency noise is greatly attenuated, and the filter is non-symmetrical. The following section details how many non-analyte filters fail in one or more of these parameters.
FILTERS A subset of digital filtering spectra is processing spectra with an order-derivative, such as a zero, first, second, or higher order derivative. As demonstrated herein, derivative filters pass high frequency spectral information. Another subset of digital filtering spectra is processing spectra with a symmetrical filter, such as a Fourier filter. Both of these filter types use responses that are restricted or constrained. That is, setting a particular filter efficiency at two or more frequencies forces particular filter responses at other frequencies that is detrimental. This detrimental response is described in detail in the subsequent paragraphs. Analyte filters, as defined herein, do not carry the restraint of low degrees of freedom, of forcing a magnitude of a frequency response at an «th frequency based upon the magnitude of the frequency response at a first, second, third, or fourth frequency, or being forced to be symmetrical.
Order Derivative Filters
Referring now to Figure 7, several examples of zero-order derivative filters 600, presented in frequency space, are provided. The three represented filters are all Savitsky-Golay zero-order derivative filters with 5, 15, and 35-point windows, respectively. The 5-point zero-order derivative filter 601 is observed to have a high response at low frequency and to generally roll off at higher frequency. The response of this filter is to pass low frequency information, such as baseline deviations, and to remove some high frequency information from the spectra, such as noise. The 15-point zero-order derivative filter 602 again passes very low frequency information but has a much steeper roll-off. The steeper roll off results in a greater percentage removal of semi-high frequency noise. Notably, the filter rings at higher frequency. Each of the local maxima at higher frequency result in some high frequency noise being passed by the filter. These local maxima are referred to as leaks in the filter. The 35-point zero-order derivative filter 603 again passes low frequency information and has a sharper cut-off. Generally, a zero-order derivative filter passes low frequency information, reduces high-frequency information, has a steeper negative slope with an increasing number of points, and leaks some high frequency response. All of these filters leak high-frequency noise and defining two points of a filter defines the magnitude at all frequencies for a given filter. Referring now to Figure 8, several examples of first-order derivative filters 700, presented in frequency space, are provided. The three represented filters are all Savitsky-Golay first-derivative filters with 5, 15, and 35-point windows, respectively. The 5-point first-order derivative filter 701 strongly attenuates very low frequencies, is observed to have a peak response at a low-frequency, again attenuates to a degree at higher frequency, and has ripple at still greater frequency. The response of this filter is thus to attenuate very low frequency information greatly, such as baseline deviations, to pass low-frequency, and to again greatly attenuate high frequency noise with secondary ripple passing high frequency noise. The 15-point first-order derivative filter 702 attenuates very low frequency information but has a narrower passband at lower frequency than the 5-point first-order derivative filter, and more ripples leaking high frequency noise. The 35-point first-order derivative filter 703 attenuates to a degree low and high-frequency information, has a passband at relatively lower frequency, and has many ripples leaking higher frequency information, such as high-frequency noise. Generally, a first-order derivative filter filters both high and low frequency information, has a passband that passes narrower regions at lower frequencies with an increase in the number of filter points in the convolution, leaks high frequency response through the ripples of the filter, and has a low degree of freedom.
Referring now to Figure 9, several examples of second-order derivative filters 800, presented in frequency space, are provided. The three represented filters are all Savitsky-Golay second-derivative filters with 5, 15, and 35-point windows, respectively. The 5-point second-derivative filter 801 greatly attenuates low frequency, is observed to have a peak response at an intermediate frequency, again attenuates to a degree at higher frequency, and has ripple at still greater frequency. Thus, the response of this filter is to attenuate low frequency information greatly, such as baseline deviations, to pass intermediate-frequency, and to attenuate high frequency noise somewhat with secondary ripple passing high frequency noise. The 15-point second derivative filter 802 again greatly attenuates very low frequency information, but has a narrower passband at lower frequency than the 5-point second-derivative filter, and more ripples leaking high frequency noise. The 35-point second derivative filter 803 attenuates low and high-frequency information, has a passband at relatively lower frequency, and has many ripples leaking higher frequency information, such as high-frequency noise. Generally, a second-order derivative filter filters both high and low frequency information, has a passband that passes narrower regions at lower frequency with an increase in the number of filter points in the convolution, leaks high frequency response through the ripples of the filter, and has a low degree of freedom.
Higher order derivative filters have responses that are extended trends from the first and second derivative filters. Generally, a higher order derivative filter filters both high and low frequency information, has a passband that passes narrower regions at lower frequency with an increase in the number of filter points in the convolution, leaks high frequency response through the ripples of the filter, and is restricted in terms of magnitude response across frequency.
Order-filters are restricted in terms of independent magnitude of response versus frequency. For example, if a desired filtering efficiency is set at two frequencies, then an order-filter that achieves this response necessarily has some filtering efficiency at another frequency. For example, forcing a filter to greatly attenuate very low frequencies results in ripples passing higher frequency. As a second example, forcing a narrower bandpass results in a greater number of ripples. The restriction of one shape of the filter by defining another limits performance of the filters. Analyte filters are not limited by these restrictions.
Fourier Filters Another example of a filter is a Gaussian filter. A Gaussian filter reduces both baseline deviations occurring at low normalized frequency units, reduces high frequency noise occurring at high normalized frequency units, and has a passband over a range of intermediate normalized digital frequency units. A Gaussian function is defined by its mean and standard deviation. This restricts the curvature of the of cut-on and the cut-off of the filter to be symmetrical. This symmetry restricts the flexibility of having a sharper cut-on versus the cut-off or vise versa. Referring now to Figure 10, an example of a Gaussian filter is provided having a mean of 0.2 and a standard deviation of 0.05. Fourier filters operate by performing a Fourier transform of an absorbance spectrum as a function of wavelength to represent the data in intensity versus frequency. The transformed spectrum multiplied by a Gaussian function and an inverse Fourier transform is subsequently performed to transform the spectrum back to its original domain. Again, this Gaussian filter type is constrained as a Gaussian filter is defined by a mean and a standard deviation so that the roll-off of one side of the filter necessarily impacts the roll-off of the opposite side of the filter. A Gaussian filter is restricted. If two points are defined, then the rest of the Gaussian filter shape is defined. The low degree of freedom restriction of shape and/or symmetry of a Gaussian filter is not present in an analyte filter. This allows the analyte filter to be designed or built based upon the actual frequency response of the analyte in the presence of noise and interference.
Analyte Filters
An analyte filter is a filter that preferably rejects variations that are likely to be detrimental to the measurement system, while passing the signal that is most likely unique to the target analyte. For example, an analyte filter based upon shape information of the target analyte, which is based upon the analyte signal in the presence of noise and/or interference, is represented by the net analyte signal, or is based upon one or more extracted features. A typical use of an analyte filter is to improve the robustness of a calibration with respect to sources of interference. Typically, an analyte filter is built or optimized to a set of specifications that can include: the analyte signal, noise, and interference. Preferably an analyte filter is used in combination with a multivariate soft model, such as partial least square or principal component regression, as soft models have difficulties compensating for noise, baseline variation, and interferences. The analyte filter removes, minimizes, or reduces noise, baseline deviation, and interferences.
To enhance the analyte filter functionality, the analyte filter is optionally implemented in several stages, each stage designed and optimized according to a particular criterion. In one case, an analyte filter is generated or enhanced through a theoretical design that involves the use of interference models, in addition to shape information, to develop an optimized analyte filter. In a second case, an iterative approach is used to generate or enhance an analyte filter shape, where the iterative approach uses knowledge of the analyte and/or noise frequency response of the system. An analyte filter is not restricted in the frequency domain in the manner of a derivative leaking excess high frequency noise, is not restricted to a symmetrical shape, and has an n" point in the frequency domain not restricted by a combination of two, three, four, five or more points of the analyte filter in the frequency domain.
Based upon the specification and/or design approach of an analyte filter, a number of approaches are used to generate an analyte filter for an application. Several example processes of analyte filter design are provided, infra.
Example Il
Shape Based Design
An analyte filter is optionally created or optimized using the shape of the target analyte, preferably in conjunction with interference. The pure component spectrum of a target analyte is often considered the signal of interest in applications involving spectroscopy and noninvasive analyte property determination. The use of preprocessing techniques is traditionally aimed at enhancing and extracting the pure component analyte signal. However, the true signal of interest in multivariate applications is contained in the shape of the target analyte after orthogonalization with respect to the interference. Specifically, the net analyte signal defines that portion of the pure component analyte spectrum that is accessible for analyte measurement and is often significantly different than the pure component spectrum related to the target analyte.
Therefore, shape is the signal in the presence of interference. One measure of signal strength is the net analyte signal, nas, given by the projection of the pure component analyte signal, R, onto the null space of the interference contained in the B, where / is the identity matrix as provided in equation 1.
Figure imgf000019_0001
The interference, B, is a basis set that spans at least a portion of the space of expected variation. Alternately, nas is estimated from the regression vector of the multivariate calibration.
The net analyte signal constitutes a pattern, contained in the vector nas, with shape information that is used to determine the optimal filter characteristics. The frequency response of the nas, denoted nasφ, provides the magnitude and phase of the signal of interest versus the normalized frequency and is used to establish filter criteria yielding minimal attenuation of the signal of interest. In particular, a filter is designed to pass frequency components of the nas that are significant while rejecting harmonics that contribute minimally to the analyte property estimation.
Several applicable cases of the invention are provided as exemplar embodiments. In one case, a primary interference affecting the analyte measurement is related to low frequency harmonics that develop due to instrument drift, instrument-to- instrument differences, and/or gross changes in the scattering properties of the tissue between subjects and over time. The enhancement of analytical signal, optionally measured by the nαs, is performed by rejecting low frequency interference and passing frequency components that has the highest probability of representing the signal of interest in subsequent samples.
Therefore, a high-pass filter is implemented with a break frequency or cut-on,^., that is established on the basis of nαsφ. The selection of break frequency is based upon the point at which \nαs(f)\ falls below a minimum threshold level, which is typically about an order of magnitude less than the maximum frequency component though of fractional order of magnitude or two or more orders of magnitude less thresholds is optionally used. In addition, specific features of the analyte signal, the signal in the presence of interference, or the nas are extracted to provide further information to enhance the design of the analyte filter. For example, the most influential regions of the nas are characterized according to wavelength range, frequency content, and bandwidth. Optionally, these features are extracted using preprocessing techniques, described infra. Optimal filters are employed as a function of the influential individual regions yielding a net improvement in the analyte performance.
Example III Theoretical Design
In another embodiment of the invention, a-priori knowledge of the measured system is used in the creation and/or optimization of an analyte filter. The use of a-priori knowledge for the development of an analyte filter is accomplished by selecting, parameterizing, and implementing an optimized filter using what is known about noise, interference, and the signal of interest. This method does not directly rely on a set of empirical data for the determination of the analyte filter, but instead uses theoretical methods to make a robust and optimal system for separating the spectroscopic signal from noise and interference. Optionally, a-priori knowledge is used in conjunction with, subsequent to, concurrently with, or iteratively with empirical data in analyte filter creation.
The a-prior information about random noise is in the form of one or more models, such as a statistical model, that are used to provide the corresponding frequency characteristics, N(f) . For example, the a-priori information about interfering analytes is provided through an estimate of the net analyte signal, extracted features, interferences, and/or noise characteristics, which provide shape information about the signal of interest. A-priori knowledge in analyte filter design is optionally used in combination with preprocessing.
There are two fundamental steps for the theoretical design. First, the low frequency variation over the wavelength axis of the measured spectrum is attenuated through a high-pass filtering operation. The cut-off frequency or bandwidth of the high-pass operation is determined according to the shape information of the nas as previously discussed. The second function is a low-pass or smoothing operation in which noise is suppressed through the attenuation of high frequencies. The break frequency of the low-pass filter is set according to the bandwidth of the spectrometer, the net analyte signal of the target analyte or constituent, and/or the necessary signal-to- noise ratio required to make the measurement. For example, a-priori knowledge of the system is used to generate a two-stage filter.
In one embodiment of the invention, the measured spectrum is oversampled with respect to the wavelength axis and the low-pass bandwidth is set based upon an analysis of the signal-to-noise ratio where the net analyte signal is the signal and the noise is the root-mean-square variation of the measured spectrum in the wavelength region used for measurement of the target analyte. As the low-pass bandwidth is reduced, the high frequency components of the noise are attenuated, leading to a reduction in the root-mean-square noise. However, this process also attenuates the high frequency components of the signal leading to a simultaneous reduction in the net analyte signal. In cases in which the spectrum is oversampled with respect to the wavelength axis, the noise is distributed in greater proportions at higher frequencies than the net analyte signal. Therefore, low-pass filtering the measured spectrum removes a greater proportion of the noise than the net analyte signal. The optimal low-pass bandwidth can be defined as one that maximizes the ratio of the signal or net analyte signal to the noise. Given a frequency domain model the net analyte signal, NAS(f) , and the noise, N(f) , the bandwidth of the low-pass filter, fbw , can be determined through equation 2
Figure imgf000021_0001
as the value for fbw at which SNR(fbw) is maximized. Alternately, this can be performed iteratively by filtering a wavelength domain representation of the net analyte signal and noise at various low-pass bandwidths, or through an empirical set of data by selecting the break frequency to optimize the standard error of prediction or other figure of merit. The break frequency of the high-pass section is set to attenuate low frequency variation caused by changes in the scattering while passing the net analyte signal. This is generally accomplished empirically through an exemplary set of data or through a harmonic analysis of the net analyte signal. Optionally, signal as opposed to net analyte signal is used in this optimization.
In an alternate embodiment, the analyte filter is employed to compensate for instrument related shape distortion that causes a contraction, dilation, or shift. For example, a-prior knowledge of wavelengths of absorbance of a particular feature are optionally used as standards for a wavelength correction. In terms of an analyte filter used to adjust shift distortion, a Bessel-Thomson filter is used to provide a constant correction with respect to wavelength. The measurement of a complete spectrum enables that filter to be either causal or non-causal, which provide the opportunity to shift a spectrum in either direction along an axis. Alternatively, complex structures or adaptive means are employed for adjusting the filter characteristics on the basis of an acquired signal representative of the distortion. Adaptive means for filtering using wavelets, artificial neural networks, a Kalman filter, and related methods are also used, depending on the application.
Example IV
Array Based
In another embodiment of the invention, an analyte filter is used to take advantage of known hardware features. For example, many existing spectrometer systems use a single element detector as a means for generating a response, such as a near- infrared spectrum. A principle problem with this approach is that the serial extraction of intensity over time leads to a wavelength-to-wavelength correlation in the noise. Consequently, the frequency response of the measurement noise is dominated by low-frequency harmonics that can not be efficiently filtered without also attenuating the signal of interest. For this reason, an array-based system is employed, where a multiplicity of detectors convert spatially distributed electro-magnetic energy into a spectrum suitable for processing. Each element detects a band of wavelengths and operates independently. It is determined that because the detectors are independent, the noise versus wavelength is white and the frequency response is flat, enabling a highly efficient reduction of noise through filtering. Hence, an analyte filter optimized for the same matrix with the same chemical and environment interferences is different for an array based spectrometer and a single element detector based spectrometer. Most analytes in visible, near-infrared, or infrared spectroscopy are dominated by low frequencies. For instance, referring now to Figure 1 1 , a single element detector is characterized by an increasing magnitude of noise as the normalized frequency decreases 1 101. In stark contrast, the frequency response of an array detector is flat or nearly flat 1102. This means that the lower frequency noise of the array detector is de-emphasized. As a result, an analyte filter optimized for a glucose signal using an array detector optimally shifts to lower frequencies 1103, compared to an analyte filter optimized for a glucose signal using a single element detector 1104. Typically, molecular analytes have spectral signatures dominated by lower frequencies. A near-infrared glucose signal in frequency space 1105 exemplifies this. Accordingly, the analyte filter optimized using the flat noise characteristic moves closer to the low frequencies that dominate the glucose signal. Hence, the analyte filter generated using an array detector is different and more efficient compared to the analyte filter optimized for the glucose signal using a single element detector. The lower frequency analyte filter optimized with a detector array noise structure allows extraction of a greater portion of the glucose signal, resulting in improved performance of the analyte concentration estimation using noninvasive spectra.
Example V Cluster / Analyte Filter
When spectral interference is significant as a result of the heterogeneity, a dynamic nature of the measurement site, and/or instrument performance state or characteristics, it is beneficial to develop an analyte filter associated with a particular cluster of samples that represent a more uniform tissue state. In this particular circumstance, the analyte filter is preferably associated with a particular calibration that is specific to the variation represented by the cluster. For this purpose we disclose an adaptive filter that, in addition to the more general analyte filter, removes or attenuates spectral interference that is present in a particular cluster. The filter operates by using a basis set of background spectra, such as interference spectra, to reduce the contribution of the cluster specific interference to future measurements.
Herein, a data set is used for calibration including xc G ^PXN and yc e $R™ , where xc is the matrix of tissue measurements, yc is a vector of analyte concentrations, P is sample number, and N is data point, such as wavelength, where χc and yc are collected or associated with a particular state. A state is defined by one or more of the following: subject, day, time, tissue position, temperature, environmental condition, tissue properties, instrument state, etc. During calibration it is often advantageous to combine data associated with multiple states so that a more certain estimate of the analyte signal relative to noise is determined. However, the different states often result in highly nonlinear variation that is difficult to model. The adaptive filter uses a set of background spectra, which define the interference to optimally reduce subsequent contributions of similarly shaped interference.
The filter operates according to equation 3
Figure imgf000024_0001
where the basis set spectra, xb, the pseudo inverse, (χb τ)~~λ , and the measured spectrum, xCt are used to the corrected spectrum or analyte filtered spectrum, xc', and equation 4
1 V (4)
A=I where yc is the reference concentration, the klh element of yc \s, yc k , and L is the set of spectra used to calculated xb . During the process of measurement, a basis set is determined using the first one or more spectra associated with a particular condition. Subsequent samples are adaptively filtered using the method above until a change in the interference is detected.
The analyte filter is used to adapt a spectral reading to a nearest cluster, to translate a reading from one cluster to another cluster, and to determine outliers as described in U.S. patent application number 09/664,973 filed September 18, 2000 and in U.S. patent application number 10/870,727 filed June 16, 2004 both of which are incorporated herein in their entirety by this reference thereto.
Example Vl Multi-Stage Filter An analyte filter is optionally designed to the signal of interest compensating for noise structure without restraint on the analyte filter shape. Existing filter methodologies for noninvasive detection of analytes are limited as a consequence of the underlying filter structure. Although the design of the optimal analyte filter involves the use of multiple design goals, existing filter implementations constrain the high-pass and low-pass filter characteristics according to a deterministic function. For example, filtering involves the use of functions, such as the Savitsky-Golay derivative, which contain significant stop-band ripple and thereby pass unwanted noise. Passing unwanted high frequency noise results in performance degradation, as described supra. For example, a finite-impulse response filter is designed and implemented by constraining the frequency response according to a Gaussian curve. While the resulting step response with respect to the wavelength axis is well behaved, the low and high-pass filter characteristics are necessarily sub-optimal due to symmetry restrictions. In a second example, other methods of filtering involve the use of functions, such as the Savitsky-Golay derivative, which contain significant stop-band ripple and thereby pass unwanted noise, which leads to performance degradations. In yet another embodiment of the invention, the problems of a restrained filter or a filter leaking excess high frequency noise are addressed. The generation of a filter that is optimized according to the shape information contained in the signal of the analyte, such as the net analyte signal, and the noise is preferably performed with at least two design goals and involves a stage for each goal. For example, optimization of the signal is preferably performed with optimization of a longpass filter or with a highpass filter designed to pass the signal or net analyte signal frequency response shape. Before or after optimization of the signal, a filter is optimized for the noise structure. For example, a shortpass or lowpass filter is used or optimized based upon the noise structure so that high frequency noise is reduced, minimized, or eliminated. The signal and noise optimization functions are optionally performed in either order depending on the implementation of the band-pass filter. The methods used for performing this operation include infinite-impulse response (HR) and finite- impulse response (FIR) band-pass filtering. In the preferred embodiment a FIR filter is implemented according to equation 5,
P nij j = ∑akm (p_ψ (5)
where mf J is the filtered spectrum at the /h measured wavelength, my is the measured spectrum at the/h wavelength, ak denotes the kth filter coefficient and P is the length of the filter impulse response or filter window width. In equation 5, the filter is non-causal and applied across the wavelength axis of the measured spectrum. The filter width, P, is assumed to be an odd number and the filter coefficients are determined according to the desired filter break frequencies and characteristics of the pass and stop-bands. This multiple stage filter approach reduces and in some instances eliminates the restraints of traditional filters.
Example VII
Analyte filters are optionally built based upon specifications and/or shape. For example, combinations of longpass, shortpass and Butterworth filters are used to meet objectives set out in the specification requirements of a given design filter. Combinations of filters are created resulting in a complex filter, a two-stage filter, a three stage-filter, or an πth stage filter.
The concern of restriction of filter shape, such as with the Gaussian filters, and the concern of excessive high frequency leaking of a filter, such as with derivative filters, is avoided with non-symmetrical filters, such as shortpass filters, longpass filters, bandpass or passband filters, and MR Butterworth filters. However, these filter types are not customarily used in combination to create an analyte filter shape based upon a set of specifications. Combinations of these filters minimize high frequency leaking, reduce baseline deviation, avoid symmetry concerns, and match, correlate, or correspond to the specified shape of the analyte frequency response in the presence of interference. In tuning the combination filters to the shape of the analyte response, the order of the filters, such as the longpass filter, or shortpass filter, is varied.
Herein, a complex filter has a simple or complex shape. In general, complex filters are not restricted to the limitations of an order derivative filter or a Gaussian filter. For example, the filters need not be symmetrical and do not necessarily pass higher frequency intensity noise. A complex filter is generated in many ways. For example, a complex filter is optimized using a response function, theoretically generated using net analyte signal, designed for noise reduction, or is a complex shape designed from multiple inputs.
*
Example VIII Calibration Transfer
In still yet another embodiment of the invention, an analyte filter is generated using data from a first set of analyzers or spectrometers. An additional analyzer or a second set of analyzers that are not present in the first set of analyzers have distinct spectral characteristics that can degrade analytical performance of analyte property determination. Implementation of a model built on the first set of analyzers onto the second set of analyzers thus requires a calibration transfer or filter for optimal performance.
Overview Calibration transfer is a standardization procedure designed to eliminate a full recalibration and to maintain information residing in the existing model. Calibration transfer is useful because in the development of the training set, also referred to as a calibration set, sources of variation in the instrument, sample, and environment are modeled. Therefore, as the instrument, sample, or environment state changes the model components do not exactly match the current state. These problems are typically compounded when transferring a calibration from a first spectrometer to a second spectrometer. Calibration transfer addresses a range of concerns including: wavelength or x-axis stability, energy throughput or y-axis stability, and bandwidth. Calibration transfers are used across time on one or more instruments, across many instruments of the same design, and across instruments of different design. Calibration transfer is also used when the sample varies. A living organism, such as the human body, undergoes continuous change. In addition, calibration transfers are used to adjust for changes in the environmental conditions, such as changes in humidity or temperature. While broad models account for many changes, changing state still leads to an increase in estimation error or analyte determination.
Sources of variation
There are many potential sources of variation in an optical based analyzers, which are referred to herein as spectrometers or analyzers.
A first source of variation is when there is a change in the entire spectrometer design. This is the case, for example, when a calibration is built on a first spectrometer, such as a master spectrometer and estimations or predictions are performed on a different spectrometer, such as a slave spectrometer. This first type of variation is common when a research grade spectrometer is used for calibration and a process grade spectrometer, built with less stringent specifications, is used for subsequent analyses.
A second source of variation is a change in a part of a spectrometer. Examples of changing a part of a spectrometer include: changing a source, which can effect the blackbody radiation emitted; changing a monochromator or grating, which can effect the wavelength axis; changing a fiber optic or fiber bundle, which can effect the bandwidth; or changing the detector, which can effect the response curve or change a high or low frequency cut-off.
A third source of variation results from alignment of the spectrometer. Alignment changes result from movement of one or more of sub-components of the spectrometer. Alignment issues also result from mechanical design and fabrication tolerances.
A fourth source of variation is change in the environmental conditions about the spectrometer or analyzer. Common environmental changes affecting near-infrared spectra include temperature and humidity.
A fifth source of variation is a change in the sample. Examples of changes in the sample include changes in chemical composition, sample finish, particle size distribution, and density. Examples of changes in a skin sample include changes in temperature, chemical composition, chemical distribution, physical structure, hydration, and/or localized pressure.
In its broadest sense, the state of the spectrometer affects its output. Variation in state results in variation of output. For example, variation in state results in variation of the observed or calculated absorption coefficient and/or the observed or calculated reduced scattering coefficient. Calibration transfer is useful for adjusting for or compensating for the change in output resulting from change in state. Techniques
As described, supra, identical performance of analytical instruments is unrealistic even with the successful implementation of tight quality control on instrument hardware. For example, variation in the output of a source, quality of lenses or mirrors, alignment, and detector response, which are limited by manufacturing tolerances, result in differences between spectrometers even of the same design. The instrument differences result in spectra of the master instrument varying from that of the slave instrument. Variations between the spectrometers result in errors when using a calibration developed on a spectrometer to determine parameters with a second spectrometer. Generally, this error is increasingly detrimental as the signal-to-noise ratio of the determined analyte decreases.
Existing techniques, such as developing a robust model that covers all future conditions, full recalibration of an analyzer, and axis standardization using a standard are limited in terms of breadth of use, time and cost, and effectiveness.
Herein an analyte filter is used advantageously in the calibration transfer process.
The analyte filter is used to extract the signal from a known type of interference or from a particular cluster of sample as provided herein. The analyte filter is optionally used with a preprocessing and/or processing step. Several preprocessing techniques are described, infra.
Preprocessing
In data preprocessing, typically the same preprocessing is performed on data that are used to generate a calibration model and the data that the model applied to.
Generally, these techniques modify spectral response, such that spectra collected on different instruments are transformed so that the spectra appear as though they were all measured on the same instrument. However, identical preprocessing is not necessary. Preprocessing techniques include: multiplicative scatter correction, standard normal variate transformation, finite impulse response, a Bessel filter, normalization, a moving average, a finite impulse response filter, an infinite impulse response filter, and Fourier transformation. An additional preprocessing technique is extraction of an absorbance and/or scattering signal with one or more sub-ranges of a spectrum. The selected and optionally further modified signal is preferably subsequently filtered with an analyte filter.
Although the invention is described herein with reference to the preferred embodiments, one skilled in the art will readily appreciate that other applications may be substituted for those set forth herein without departing from the spirit and scope of the present invention. Accordingly, the invention should only be limited by the Claims included below.

Claims

1. A method for determining a property of an analyte using a noninvasive spectrum, comprising the steps of: processing said noninvasive spectrum using an analyte filter yielding a processed spectrum, wherein said analyte filter correlates with frequency response shape of said analyte; and estimating said property using a multivariate model and said processed spectrum.
2. The method of Claim 1 , further comprising the step of: extracting a feature representative of said response shape of said analyte in frequency response.
3. The method of Claim 2, wherein said analyte filter enhances said shape in frequency response below a break frequency.
4. The method of Claim 3, wherein said analyte filter enhances said shape in frequency responses above a break frequency.
5. The method of Claim 2, wherein said feature is characterized in terms of signal and interference.
6. The method of Claim 1 , wherein said multivariate model comprises any of: principal component regression; and partial least squares.
7. The method of Claim 1 , wherein said frequency response shape of said analyte comprises any of: a net analyte signal; an accessible portion of a pure component analyte spectrum; and shape of said analyte after orthogonalization with respect to interference.
8. The method of Claim 7, further comprising the step of: calculating said shape of said net analyte signal according to:
Figure imgf000033_0001
wherein said net analyte signal, nas, is given by the projection of the pure component analyte signal, R, onto the null space of the interference contained in B, wherein B is a basis set that spans at least a portion of the space of expected variation and / is the identity matrix.
9. The method of Claim 7, further comprising the step of: estimating said net analyte signal from a regression vector of said multivariate model.
10. The method of Claim 1 , further comprising the step of: theoretically generating said analyte filter using frequency response shape of said analyte.
11. The method of Claim 10, wherein said step of theoretically generating further comprises the steps of: applying a-pήori knowledge of signal of a system from which said noninvasive spectrum is collected; and applying a-priori knowledge of interference of said system from which said noninvasive spectrum is collected.
12. The method of Claim 11 , wherein said step of theoretically generating further comprises the step of: generating a multi-stage filter.
13. The method of Claim 12, wherein said multi-stage filter comprises: a two-stage filter.
14. The method of Claim 13, wherein said two-stage filter comprises: a lowpass filter; and a highpass filter.
15. The method of Claim 1 , further comprising the step of: preprocessing said noninvasive spectrum prior to said step of processing, wherein preprocessing enhances at least a portion of said shape of said analyte.
16. The method of Claim 10, further comprising the step of: iteratively generating frequency response of said analyte filter.
17. The method of Claim 16, wherein said steps of theoretically generating and iteratively generating comprise any of: sequential performance; concurrent performance; and iterative performance.
18. The method of Claim 1 , further comprising the step of: iteratively generating frequency response shape of said analyte filter using a set of data from a first set of spectrometers; and implementing said analyte filter on a second spectrometer, wherein said second spectrometer is not a member of said first set of spectrometers.
19. The method of Claim 1 , further comprising the step of: generating said analyte filter using a set of spectra from a first set of spectrometers.
20. The method of Claim 19, further comprising the step of: implementing said analyte filter on at least one spectrum from a second set of spectrometers, wherein a union of said first set of spectrometers and said second set of spectrometers results in an empty set.
21. The method of Claim 20, wherein said step of estimating further comprises the step of: performing multivariate analysis.
22. The method of Claim 21 , where said step of performing multivariate analysis comprises use of any of: a regression model; partial least squares; and a principal component regression.
23. The method of Claim 1 , wherein said analyte filter comprises an 7?th point in the frequency domain not restricted by said analyte filter points at a combination of a first, a second, a third, and/or a fourth point of the analyte filter in the frequency domain.
24. The method of Claim 1 , wherein said analyte filter comprises: a cluster filter.
25. The method of Claim 24, further comprising any of the steps of: attenuating spectral interference present in a cluster; and removing spectral interference present in said cluster.
26. The method of Claim 25, wherein said cluster filter facilitates calibration transfer of said noninvasive spectrum.
27. The method of Claim 25, wherein said cluster filter compensates for variation in any of: a sample; an analyzer; and an environmental state.
28. The method of Claim 25, wherein said multivariate model is associated with said cluster.
29. The method of Claim 25, wherein said cluster filter compensates for variation of: a subject; a sampling position; temperature; and instrument state.
30. The method of Claim 25, wherein said cluster filter operates according to:
Figure imgf000037_0001
wherein the basis set spectra, xb, a pseudo inverse, (χ[)"! , and a measured spectrum, xCt are used to the correct said noninvasive spectrum, xc'.
31. The method of Claim 26, wherein said cluster filter operates according to:
Figure imgf000037_0002
wherein yc is a reference concentration, a kih element of yc \s, yc k , and L is a set of spectra used to calculated x. .
32. The method of Claim 1 , wherein said analyte filter comprises a product of a derivative order filter and a shortpass filter, wherein high frequency noise leaked by said derivative order filter is reduced.
33. The method of Claim 1 , further comprising the step of: developing said analyte filter using said analyte property shape.
34. The method of Claim 33, further comprising the step of: using interference of said analyte in said step of developing said analyte filter.
35. The method of Claim 1 , wherein said analyte filter comprises a multi-stage filter.
36. The method of Claim 35, wherein said multi-stage filter comprises a two-stage filter.
37. The method of Claim 1 , wherein said noninvasive spectrum comprises: a near-infrared spectrum; at least a portion of said near-infrared spectrum being in the range 1150 to 1850 nm.
38. The method of Claim 37, wherein said analyte property comprises: a glucose concentration.
39. The method of Claim 38, further comprising the step of: generating said noninvasive spectrum using a photo diode array detector.
40. The method of Claim 39, further comprising the step of: implementing said analyte filter on an analyzer not used to develop said analyte filter.
41. The method of Claim 40, wherein said analyte filter passes an average of less than 0.05 magnitude over a normalized frequency range of 0.2 to 0.5.
42. The method of Claim 41 , further comprising the step of: preprocessing said noninvasive spectrum.
43. The method of Claim 42, wherein said step of preprocessing comprises use of any of: multiplicative scatter correction; a standard normal variate transformation; a finite impulse response; a Bessel filter; normalization; a moving average; a finite impulse response filter; a Bessel function; an infinite impulse response filter; and a Fourier transformation.
44. An apparatus for determining an analyte property using a noninvasive spectrum, comprising: means for obtaining said noninvasive spectrum; and an analyzer, comprising: a processor; an algorithm expanded by said processor; an analyte filter operating as a step in said algorithm; and a multivariate model operating as a step in said algorithm, said analyzer operating on said noninvasive spectrum using said analyte filter to yield a processed spectrum; said analyte filter correlating with frequency response shape of said analyte; and said multivariate model operating on said processed spectrum to yield said analyte property.
45. The apparatus of Claim 44, wherein said shape comprises an extracted feature of said analyte in frequency response.
46. The apparatus of Claim 45, wherein said algorithm further comprises: preprocessing means for extracting said shape of said analyte.
47. The apparatus of Claim 44, wherein said analyte filter enhances shape in frequency response below a break frequency.
48. The apparatus of Claim 47, wherein said analyte filter enhances shape in frequency responses above a break frequency.
49. The apparatus of Claim 44, wherein said extracted feature is characterized in terms of signal and interference.
50. The apparatus of Claim 44, wherein said frequency response shape of said analyte comprises any of: a net analyte signal; an accessible portion of a pure component analyte spectrum; and shape of said analyte after orthogonalization with respect to interference.
51. The apparatus of Claim 50, wherein said shape comprises said net analyte signal according to: nas = [l- BB+ ]R.
wherein said net analyte signal, nas, is given by the projection of the pure component analyte signal, R, onto the null space of the interference contained in B, wherein B is a basis set that spans at least a portion of the space of expected variation and / is the identity matrix.
52. The apparatus of Claim 50, wherein said net analyte signal is estimated from a regression vector of said multivariate model.
53. The apparatus of Claim 44, wherein said analyte filter comprises: a theoretically generated filter using frequency response shape of said analyte.
54. The apparatus of Claim 53, wherein said theoretically generated filter uses: a-priori knowledge of signal of a system from which said noninvasive spectrum is collected; and a-priori knowledge of interference of said system from which said noninvasive spectrum is collected.
55. The apparatus of Claim 54, wherein said theoretically generated filter comprises: a multi-stage filter.
56. The apparatus of Claim 55, wherein said multi-stage filter comprises: a two-stage filter.
57. The apparatus of Claim 56, wherein said two-stage filter comprises: a lowpass filter; and a highpass filter.
58. The apparatus of Claim 56, wherein said algorithm further comprises: means for preprocessing said noninvasive spectrum prior to said step of processing.
59. The apparatus of Claim 53, wherein said analyte filter is iteratively generated using frequency response shape of said analyte.
60. The apparatus of Claim 44, wherein said analyte filter is generated using a set of data from a set of spectrometers, wherein said analyzer is not a member of said set of spectrometers.
61. The method of Claim 44, where said multivariate model comprises any of: a regression model; partial least squares; and principal component regression.
62. The apparatus of Claim 44, wherein said analyte filter comprises an πth point in the frequency domain not restricted by said analyte filter points at a combination of a first, a second, a third, and/or a fourth point of the analyte filter in the frequency domain.
63. The apparatus of Claim 44, wherein said analyte filter comprises: a cluster filter.
64. The apparatus of Claim 63, wherein said cluster filter either removes spectral interference or attenuates spectral interference present in a cluster.
65. The apparatus of Claim 63, wherein said cluster filter facilitates calibration transfer of said noninvasive spectrum.
66. The apparatus of Claim 63, wherein said cluster filter is used to compensate for variation in any of: a sample; a subject; a sampling position; temperature; instrument state; an analyzer; and an environmental state.
67. The apparatus of Claim 63, wherein multivariate model is associated with said cluster filter.
68. The apparatus of Claim 63, wherein said cluster filter operates according to:
Figure imgf000043_0001
wherein a basis set spectra, xb, a pseudo inverse, (χ[)"' , and a measured spectrum, Xc, are used to generate a corrected noninvasive spectrum, χc'.
69. The apparatus of Claim 68, wherein said cluster filter operates according to:
Figure imgf000043_0002
wherein yc is a reference concentration, a k[h element of yc is, yc l , and L is a set of spectra used to calculated xb .
70. The apparatus of Claim 44, wherein said analyte filter comprises: a product of a derivative order filter and a shortpass filter, wherein high frequency noise leaked by said derivative order filter is reduced.
71. The apparatus of Claim 44, further comprising: a photo diode array detector.
72. The apparatus of Claim 44, wherein said analyte filter is optimized using said shape.
73. The apparatus of Claim 66, wherein said shape comprises: a roll-on frequency; a roll-off frequency; signal; and noise.
74. The apparatus of Claim 73, wherein said noise comprises a homogeneous function in frequency space.
75. The apparatus of Claim 72, wherein said analyte filter comprises: a multi-stage filter.
76. The apparatus of Claim 75, wherein said multi-stage filter comprises: a two-stage filter.
77. The apparatus of Claim 72, wherein said analyte filter is generated theoretically.
78. The apparatus of Claim 72, wherein said filter comprises: average attenuation of less than 0.05 magnitude above 0.25 frequency units.
79. The apparatus of Claim 78, wherein said filter comprises: a maximum gain of less than 0.1 magnitude above 0.25 normalized frequency units.
80. The apparatus of Claim 79, wherein said average attenuation comprises a value less than 0.025 magnitude and said maximum gain comprises a value less than 0.05 magnitude.
81. The apparatus of Claim 72, wherein said analyzer comprises: a sample module in a first housing; and a base module in a second housing, wherein said sample module communicates with said base module.
82. The apparatus of Claim 72, wherein said analyzer comprises: a photo-diode array.
83. The apparatus of Claim 72, wherein said analyzer further comprises: a coupling fluid; and a guide to acquire, wherein said coupling fluid and said guide couple said sample module to a tissue sample during use.
84. The apparatus of Claim 72, wherein said analyte filter is generated with a set of spectrometers, wherein said set of spectrometers does not comprise said analyzer.
PCT/US2005/027818 2004-08-06 2005-08-05 Analyte filter method and apparatus WO2006017700A2 (en)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US59943104P 2004-08-06 2004-08-06
US60/599,431 2004-08-06
US11/188,064 US7436511B2 (en) 1999-01-22 2005-07-22 Analyte filter method and apparatus
US11/188,064 2005-07-22

Publications (2)

Publication Number Publication Date
WO2006017700A2 true WO2006017700A2 (en) 2006-02-16
WO2006017700A3 WO2006017700A3 (en) 2007-06-07

Family

ID=35839929

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2005/027818 WO2006017700A2 (en) 2004-08-06 2005-08-05 Analyte filter method and apparatus

Country Status (2)

Country Link
US (1) US7436511B2 (en)
WO (1) WO2006017700A2 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102519944B (en) * 2006-03-10 2015-04-01 康宁股份有限公司 Optimized method for lid biosensor resonance detection
CN108226064A (en) * 2016-12-15 2018-06-29 株式会社堀场制作所 Analytical equipment, computer readable storage medium and analysis method

Families Citing this family (143)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2501825C (en) 2002-10-09 2009-12-01 Therasense, Inc. Fluid delivery device, system and method
US7727181B2 (en) 2002-10-09 2010-06-01 Abbott Diabetes Care Inc. Fluid delivery device with autocalibration
US7993108B2 (en) 2002-10-09 2011-08-09 Abbott Diabetes Care Inc. Variable volume, shape memory actuated insulin dispensing pump
US7679407B2 (en) * 2003-04-28 2010-03-16 Abbott Diabetes Care Inc. Method and apparatus for providing peak detection circuitry for data communication systems
US20190357827A1 (en) 2003-08-01 2019-11-28 Dexcom, Inc. Analyte sensor
US7299082B2 (en) * 2003-10-31 2007-11-20 Abbott Diabetes Care, Inc. Method of calibrating an analyte-measurement device, and associated methods, devices and systems
US20070135697A1 (en) * 2004-04-19 2007-06-14 Therasense, Inc. Method and apparatus for providing sensor guard for data monitoring and detection systems
US9636450B2 (en) 2007-02-19 2017-05-02 Udo Hoss Pump system modular components for delivering medication and analyte sensing at seperate insertion sites
US20100331646A1 (en) * 2009-06-30 2010-12-30 Abbott Diabetes Care Inc. Health Management Devices and Methods
US7547281B2 (en) 2005-02-01 2009-06-16 Medtronic Minimed, Inc. Algorithm sensor augmented bolus estimator for semi-closed loop infusion system
EP1863559A4 (en) * 2005-03-21 2008-07-30 Abbott Diabetes Care Inc Method and system for providing integrated medication infusion and analyte monitoring system
US7768408B2 (en) 2005-05-17 2010-08-03 Abbott Diabetes Care Inc. Method and system for providing data management in data monitoring system
US7620437B2 (en) 2005-06-03 2009-11-17 Abbott Diabetes Care Inc. Method and apparatus for providing rechargeable power in data monitoring and management systems
CA2620586A1 (en) 2005-08-31 2007-03-08 Boris P. Kovatchev Improving the accuracy of continuous glucose sensors
US7756561B2 (en) * 2005-09-30 2010-07-13 Abbott Diabetes Care Inc. Method and apparatus for providing rechargeable power in data monitoring and management systems
US8880138B2 (en) 2005-09-30 2014-11-04 Abbott Diabetes Care Inc. Device for channeling fluid and methods of use
US7583190B2 (en) * 2005-10-31 2009-09-01 Abbott Diabetes Care Inc. Method and apparatus for providing data communication in data monitoring and management systems
US7736310B2 (en) 2006-01-30 2010-06-15 Abbott Diabetes Care Inc. On-body medical device securement
EP1996910A2 (en) 2006-02-22 2008-12-03 Vivum Nexus LLC Method and device for analyte measurement
US7981034B2 (en) 2006-02-28 2011-07-19 Abbott Diabetes Care Inc. Smart messages and alerts for an infusion delivery and management system
US7885698B2 (en) 2006-02-28 2011-02-08 Abbott Diabetes Care Inc. Method and system for providing continuous calibration of implantable analyte sensors
US7826879B2 (en) 2006-02-28 2010-11-02 Abbott Diabetes Care Inc. Analyte sensors and methods of use
US8374668B1 (en) 2007-10-23 2013-02-12 Abbott Diabetes Care Inc. Analyte sensor with lag compensation
US9675290B2 (en) 2012-10-30 2017-06-13 Abbott Diabetes Care Inc. Sensitivity calibration of in vivo sensors used to measure analyte concentration
US9339217B2 (en) 2011-11-25 2016-05-17 Abbott Diabetes Care Inc. Analyte monitoring system and methods of use
US7653425B2 (en) 2006-08-09 2010-01-26 Abbott Diabetes Care Inc. Method and system for providing calibration of an analyte sensor in an analyte monitoring system
US9392969B2 (en) 2008-08-31 2016-07-19 Abbott Diabetes Care Inc. Closed loop control and signal attenuation detection
US8346335B2 (en) 2008-03-28 2013-01-01 Abbott Diabetes Care Inc. Analyte sensor calibration management
US8473022B2 (en) 2008-01-31 2013-06-25 Abbott Diabetes Care Inc. Analyte sensor with time lag compensation
US8219173B2 (en) 2008-09-30 2012-07-10 Abbott Diabetes Care Inc. Optimizing analyte sensor calibration
US9326709B2 (en) * 2010-03-10 2016-05-03 Abbott Diabetes Care Inc. Systems, devices and methods for managing glucose levels
US8140312B2 (en) 2007-05-14 2012-03-20 Abbott Diabetes Care Inc. Method and system for determining analyte levels
US7630748B2 (en) 2006-10-25 2009-12-08 Abbott Diabetes Care Inc. Method and system for providing analyte monitoring
US8224415B2 (en) 2009-01-29 2012-07-17 Abbott Diabetes Care Inc. Method and device for providing offset model based calibration for analyte sensor
US7618369B2 (en) 2006-10-02 2009-11-17 Abbott Diabetes Care Inc. Method and system for dynamically updating calibration parameters for an analyte sensor
US8206296B2 (en) 2006-08-07 2012-06-26 Abbott Diabetes Care Inc. Method and system for providing integrated analyte monitoring and infusion system therapy management
US8932216B2 (en) 2006-08-07 2015-01-13 Abbott Diabetes Care Inc. Method and system for providing data management in integrated analyte monitoring and infusion system
US7599898B2 (en) * 2006-10-17 2009-10-06 International Business Machines Corporation Method and apparatus for improved regression modeling
AU2007308804A1 (en) 2006-10-26 2008-05-02 Abbott Diabetes Care, Inc. Method, system and computer program product for real-time detection of sensitivity decline in analyte sensors
US8579853B2 (en) * 2006-10-31 2013-11-12 Abbott Diabetes Care Inc. Infusion devices and methods
US20080161666A1 (en) * 2006-12-29 2008-07-03 Abbott Diabetes Care, Inc. Analyte devices and methods
US8930203B2 (en) * 2007-02-18 2015-01-06 Abbott Diabetes Care Inc. Multi-function analyte test device and methods therefor
CA2683959C (en) * 2007-04-14 2017-08-29 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in medical communication system
WO2009096992A1 (en) 2007-04-14 2009-08-06 Abbott Diabetes Care, Inc. Method and apparatus for providing data processing and control in medical communication system
WO2008130896A1 (en) * 2007-04-14 2008-10-30 Abbott Diabetes Care, Inc. Method and apparatus for providing data processing and control in medical communication system
US9008743B2 (en) * 2007-04-14 2015-04-14 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in medical communication system
CA2683930A1 (en) * 2007-04-14 2008-10-23 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in medical communication system
US9125548B2 (en) 2007-05-14 2015-09-08 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
US8103471B2 (en) 2007-05-14 2012-01-24 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
US20080312845A1 (en) * 2007-05-14 2008-12-18 Abbott Diabetes Care, Inc. Method and apparatus for providing data processing and control in a medical communication system
US8260558B2 (en) 2007-05-14 2012-09-04 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
US10002233B2 (en) * 2007-05-14 2018-06-19 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
US8600681B2 (en) 2007-05-14 2013-12-03 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
US8444560B2 (en) 2007-05-14 2013-05-21 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
US8239166B2 (en) 2007-05-14 2012-08-07 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
US8560038B2 (en) 2007-05-14 2013-10-15 Abbott Diabetes Care Inc. Method and apparatus for providing data processing and control in a medical communication system
WO2008144616A1 (en) * 2007-05-18 2008-11-27 Heidi Kay Lipid raft, caveolin protein, and caveolar function modulation compounds and associated synthetic and therapeutic methods
US8160900B2 (en) 2007-06-29 2012-04-17 Abbott Diabetes Care Inc. Analyte monitoring and management device and method to analyze the frequency of user interaction with the device
US8834366B2 (en) 2007-07-31 2014-09-16 Abbott Diabetes Care Inc. Method and apparatus for providing analyte sensor calibration
US20090036760A1 (en) * 2007-07-31 2009-02-05 Abbott Diabetes Care, Inc. Method and apparatus for providing data processing and control in a medical communication system
US8377031B2 (en) 2007-10-23 2013-02-19 Abbott Diabetes Care Inc. Closed loop control system with safety parameters and methods
US8409093B2 (en) 2007-10-23 2013-04-02 Abbott Diabetes Care Inc. Assessing measures of glycemic variability
US8204689B2 (en) * 2007-10-24 2012-06-19 Veris Technologies, Inc. Mobile soil mapping system for collecting soil reflectance measurements
US20090164239A1 (en) * 2007-12-19 2009-06-25 Abbott Diabetes Care, Inc. Dynamic Display Of Glucose Information
US20090164190A1 (en) * 2007-12-19 2009-06-25 Abbott Diabetes Care, Inc. Physiological condition simulation device and method
WO2009120600A2 (en) 2008-03-25 2009-10-01 The Curators Of The University Of Missouri Method and system for non-invasive blood glucose detection utilizing spectral data of one or more components other than glucose
RU2566920C2 (en) * 2008-05-22 2015-10-27 Дзе Кьюрейторз Оф Дзе Юниверсити Оф Миссури Method and system for non-invasive optic determination of blood glucose with application of data spectral analysis
US8924159B2 (en) 2008-05-30 2014-12-30 Abbott Diabetes Care Inc. Method and apparatus for providing glycemic control
US8591410B2 (en) 2008-05-30 2013-11-26 Abbott Diabetes Care Inc. Method and apparatus for providing glycemic control
CN101292875B (en) * 2008-06-06 2010-07-14 天津市先石光学技术有限公司 Method for measuring component concentration with reference wavelength
US8876755B2 (en) 2008-07-14 2014-11-04 Abbott Diabetes Care Inc. Closed loop control system interface and methods
US8734422B2 (en) 2008-08-31 2014-05-27 Abbott Diabetes Care Inc. Closed loop control with improved alarm functions
US9943644B2 (en) * 2008-08-31 2018-04-17 Abbott Diabetes Care Inc. Closed loop control with reference measurement and methods thereof
US20100057040A1 (en) 2008-08-31 2010-03-04 Abbott Diabetes Care, Inc. Robust Closed Loop Control And Methods
US8622988B2 (en) * 2008-08-31 2014-01-07 Abbott Diabetes Care Inc. Variable rate closed loop control and methods
US8986208B2 (en) 2008-09-30 2015-03-24 Abbott Diabetes Care Inc. Analyte sensor sensitivity attenuation mitigation
US7999944B2 (en) 2008-10-23 2011-08-16 Corning Incorporated Multi-channel swept wavelength optical interrogation system and method for using same
US9326707B2 (en) 2008-11-10 2016-05-03 Abbott Diabetes Care Inc. Alarm characterization for analyte monitoring devices and systems
US8103456B2 (en) 2009-01-29 2012-01-24 Abbott Diabetes Care Inc. Method and device for early signal attenuation detection using blood glucose measurements
US8560082B2 (en) 2009-01-30 2013-10-15 Abbott Diabetes Care Inc. Computerized determination of insulin pump therapy parameters using real time and retrospective data processing
DK3912551T3 (en) * 2009-02-26 2023-10-30 Abbott Diabetes Care Inc Procedure for calibrating an analyte sensor
US8497777B2 (en) * 2009-04-15 2013-07-30 Abbott Diabetes Care Inc. Analyte monitoring system having an alert
WO2010121229A1 (en) * 2009-04-16 2010-10-21 Abbott Diabetes Care Inc. Analyte sensor calibration management
US20100274515A1 (en) * 2009-04-28 2010-10-28 Abbott Diabetes Care Inc. Dynamic Analyte Sensor Calibration Based On Sensor Stability Profile
US8467972B2 (en) * 2009-04-28 2013-06-18 Abbott Diabetes Care Inc. Closed loop blood glucose control algorithm analysis
EP2425209A4 (en) 2009-04-29 2013-01-09 Abbott Diabetes Care Inc Method and system for providing real time analyte sensor calibration with retrospective backfill
EP3173014B1 (en) 2009-07-23 2021-08-18 Abbott Diabetes Care, Inc. Real time management of data relating to physiological control of glucose levels
EP3932309A1 (en) * 2009-07-23 2022-01-05 Abbott Diabetes Care, Inc. Continuous analyte measurement system
WO2011014851A1 (en) 2009-07-31 2011-02-03 Abbott Diabetes Care Inc. Method and apparatus for providing analyte monitoring system calibration accuracy
ES2912584T3 (en) 2009-08-31 2022-05-26 Abbott Diabetes Care Inc A glucose monitoring system and method
US9320461B2 (en) 2009-09-29 2016-04-26 Abbott Diabetes Care Inc. Method and apparatus for providing notification function in analyte monitoring systems
WO2011053881A1 (en) 2009-10-30 2011-05-05 Abbott Diabetes Care Inc. Method and apparatus for detecting false hypoglycemic conditions
US20110184265A1 (en) * 2010-01-22 2011-07-28 Abbott Diabetes Care Inc. Method and Apparatus for Providing Notification in Analyte Monitoring Systems
US8635046B2 (en) 2010-06-23 2014-01-21 Abbott Diabetes Care Inc. Method and system for evaluating analyte sensor response characteristics
US10092229B2 (en) 2010-06-29 2018-10-09 Abbott Diabetes Care Inc. Calibration of analyte measurement system
EP2624745A4 (en) 2010-10-07 2018-05-23 Abbott Diabetes Care, Inc. Analyte monitoring devices and methods
US20120226117A1 (en) 2010-12-01 2012-09-06 Lamego Marcelo M Handheld processing device including medical applications for minimally and non invasive glucose measurements
US10136845B2 (en) 2011-02-28 2018-11-27 Abbott Diabetes Care Inc. Devices, systems, and methods associated with analyte monitoring devices and devices incorporating the same
EP3575796B1 (en) 2011-04-15 2020-11-11 DexCom, Inc. Advanced analyte sensor calibration and error detection
US9622691B2 (en) 2011-10-31 2017-04-18 Abbott Diabetes Care Inc. Model based variable risk false glucose threshold alarm prevention mechanism
EP2783193A4 (en) 2011-11-03 2015-08-26 Verifood Ltd Low-cost spectrometry system for end-user food analysis
US8710993B2 (en) 2011-11-23 2014-04-29 Abbott Diabetes Care Inc. Mitigating single point failure of devices in an analyte monitoring system and methods thereof
US9317656B2 (en) 2011-11-23 2016-04-19 Abbott Diabetes Care Inc. Compatibility mechanisms for devices in a continuous analyte monitoring system and methods thereof
US9766126B2 (en) 2013-07-12 2017-09-19 Zyomed Corp. Dynamic radially controlled light input to a noninvasive analyzer apparatus and method of use thereof
US9351672B2 (en) 2012-07-16 2016-05-31 Timothy Ruchti Multiplexed pathlength resolved noninvasive analyzer apparatus with stacked filters and method of use thereof
US9585604B2 (en) 2012-07-16 2017-03-07 Zyomed Corp. Multiplexed pathlength resolved noninvasive analyzer apparatus with dynamic optical paths and method of use thereof
US9351671B2 (en) 2012-07-16 2016-05-31 Timothy Ruchti Multiplexed pathlength resolved noninvasive analyzer apparatus and method of use thereof
US10132793B2 (en) 2012-08-30 2018-11-20 Abbott Diabetes Care Inc. Dropout detection in continuous analyte monitoring data during data excursions
US9907492B2 (en) 2012-09-26 2018-03-06 Abbott Diabetes Care Inc. Method and apparatus for improving lag correction during in vivo measurement of analyte concentration with analyte concentration variability and range data
EP4276440A3 (en) * 2012-10-04 2024-02-21 Siemens Healthcare Diagnostics Inc. Methods and apparatus for measuring aspiration pressure at low aspiration volumes
US9474475B1 (en) 2013-03-15 2016-10-25 Abbott Diabetes Care Inc. Multi-rate analyte sensor data collection with sample rate configurable signal processing
US10433773B1 (en) 2013-03-15 2019-10-08 Abbott Diabetes Care Inc. Noise rejection methods and apparatus for sparsely sampled analyte sensor data
WO2014152034A1 (en) 2013-03-15 2014-09-25 Abbott Diabetes Care Inc. Sensor fault detection using analyte sensor data pattern comparison
GB2543655B (en) 2013-08-02 2017-11-01 Verifood Ltd Compact spectrometer comprising a diffuser, filter matrix, lens array and multiple sensor detector
WO2015102745A1 (en) 2013-12-31 2015-07-09 Abbott Diabetes Care Inc. Self-powered analyte sensor and devices using the same
CN106461461A (en) 2014-01-03 2017-02-22 威利食品有限公司 Spectrometry systems, methods, and applications
WO2015153482A1 (en) 2014-03-30 2015-10-08 Abbott Diabetes Care Inc. Method and apparatus for determining meal start and peak events in analyte monitoring systems
WO2016054079A1 (en) 2014-09-29 2016-04-07 Zyomed Corp. Systems and methods for blood glucose and other analyte detection and measurement using collision computing
CN107250739A (en) 2014-10-23 2017-10-13 威利食品有限公司 The annex of Handheld spectrometer
WO2016125165A2 (en) 2015-02-05 2016-08-11 Verifood, Ltd. Spectrometry system with visible aiming beam
WO2016125164A2 (en) 2015-02-05 2016-08-11 Verifood, Ltd. Spectrometry system applications
US9983693B2 (en) * 2015-03-13 2018-05-29 Adtile Technologies Inc. Spatial motion-based user interactivity
WO2016162865A1 (en) 2015-04-07 2016-10-13 Verifood, Ltd. Detector for spectrometry system
US10066990B2 (en) 2015-07-09 2018-09-04 Verifood, Ltd. Spatially variable filter systems and methods
US11553883B2 (en) 2015-07-10 2023-01-17 Abbott Diabetes Care Inc. System, device and method of dynamic glucose profile response to physiological parameters
US9891164B2 (en) * 2015-08-27 2018-02-13 Honeywell Limited Holmium oxide glasses as calibration standards for near infrared moisture sensors
US10203246B2 (en) 2015-11-20 2019-02-12 Verifood, Ltd. Systems and methods for calibration of a handheld spectrometer
US10024780B2 (en) * 2015-12-18 2018-07-17 Abbott Laboratories (Diagnostics Division) Methods for detecting events in a flow cytometer
US9554738B1 (en) 2016-03-30 2017-01-31 Zyomed Corp. Spectroscopic tomography systems and methods for noninvasive detection and measurement of analytes using collision computing
US10254215B2 (en) 2016-04-07 2019-04-09 Verifood, Ltd. Spectrometry system applications
EP3488204A4 (en) 2016-07-20 2020-07-22 Verifood Ltd. Accessories for handheld spectrometer
US10791933B2 (en) 2016-07-27 2020-10-06 Verifood, Ltd. Spectrometry systems, methods, and applications
US11596330B2 (en) 2017-03-21 2023-03-07 Abbott Diabetes Care Inc. Methods, devices and system for providing diabetic condition diagnosis and therapy
CN111246797A (en) 2017-10-24 2020-06-05 德克斯康公司 Pre-attached analyte sensors
US11331022B2 (en) 2017-10-24 2022-05-17 Dexcom, Inc. Pre-connected analyte sensors
US11141083B2 (en) 2017-11-29 2021-10-12 Samsung Electronics Co., Ltd. System and method for obtaining blood glucose concentration using temporal independent component analysis (ICA)
KR102588906B1 (en) 2017-12-01 2023-10-13 삼성전자주식회사 Apparatus and method for assessment biological signal quality
US11457872B2 (en) 2017-12-01 2022-10-04 Samsung Electronics Co., Ltd. Bio-signal quality assessment apparatus and bio-signal quality assessment method
KR20200100997A (en) 2019-02-19 2020-08-27 삼성전자주식회사 Apparatus and method for estimating analyte concentration
CN109682976B (en) * 2019-02-28 2020-05-22 东北大学 Continuous blood glucose monitoring sensor online fault detection method based on multi-model fusion
BR112020025314A2 (en) * 2019-06-04 2021-03-09 Waveform Technologies, Inc. SENSOR SIGNAL PROCESSING WITH KALMAN-BASED CALIBRATION
WO2021020986A1 (en) * 2019-07-31 2021-02-04 Schlumberger Canada Limited A method of determining depths of wellbore reflectors
CN114280180B (en) * 2021-12-21 2023-11-14 山西大学 Method for accurately detecting types and contents of components in Chinese patent medicine in real time

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6601582B2 (en) * 1997-06-10 2003-08-05 Smithkline Beecham Corporation Dispenser with doses' counter
WO2003063699A1 (en) * 2002-01-25 2003-08-07 Sensys Medical, Inc. Indirect measurement of tissue analytes through tissue properties
WO2004000113A1 (en) * 2002-06-25 2003-12-31 Sensys Medical, Inc. Targeted interference subtraction applied to near-infrared measurement of analytes
US20050240090A1 (en) * 2003-03-07 2005-10-27 Ruchti Timothy L Method of processing noninvasive spectra

Family Cites Families (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5204532A (en) * 1989-01-19 1993-04-20 Futrex, Inc. Method for providing general calibration for near infrared instruments for measurement of blood glucose
WO1991002975A1 (en) * 1989-08-21 1991-03-07 The Board Of Regents Of The University Of Washington Multiple-probe diagnostic sensor
US5459677A (en) * 1990-10-09 1995-10-17 Board Of Regents Of The University Of Washington Calibration transfer for analytical instruments
MY107650A (en) * 1990-10-12 1996-05-30 Exxon Res & Engineering Company Method of estimating property and / or composition data of a test sample
US5243546A (en) * 1991-01-10 1993-09-07 Ashland Oil, Inc. Spectroscopic instrument calibration
US5424545A (en) * 1992-07-15 1995-06-13 Myron J. Block Non-invasive non-spectrophotometric infrared measurement of blood analyte concentrations
US5436309A (en) * 1993-04-19 1995-07-25 Dow Corning Corporation Photo-crosslinkable silicones comprising oxyalkylene styrene groups and their use
US5596992A (en) * 1993-06-30 1997-01-28 Sandia Corporation Multivariate classification of infrared spectra of cell and tissue samples
US5435309A (en) * 1993-08-10 1995-07-25 Thomas; Edward V. Systematic wavelength selection for improved multivariate spectral analysis
US5553616A (en) * 1993-11-30 1996-09-10 Florida Institute Of Technology Determination of concentrations of biological substances using raman spectroscopy and artificial neural network discriminator
US5459317A (en) * 1994-02-14 1995-10-17 Ohio University Method and apparatus for non-invasive detection of physiological chemicals, particularly glucose
US5710713A (en) * 1995-03-20 1998-01-20 The Dow Chemical Company Method of creating standardized spectral libraries for enhanced library searching
FR2734360B1 (en) * 1995-05-19 1997-07-04 Elf Antar France METHOD OF CORRECTING A SIGNAL DELIVERED BY A MEASURING INSTRUMENT
US5606164A (en) * 1996-01-16 1997-02-25 Boehringer Mannheim Corporation Method and apparatus for biological fluid analyte concentration measurement using generalized distance outlier detection
US6031232A (en) * 1995-11-13 2000-02-29 Bio-Rad Laboratories, Inc. Method for the detection of malignant and premalignant stages of cervical cancer
US5725480A (en) * 1996-03-06 1998-03-10 Abbott Laboratories Non-invasive calibration and categorization of individuals for subsequent non-invasive detection of biological compounds
US5945675A (en) * 1996-03-18 1999-08-31 Pacific Northwest Research Foundation Methods of screening for a tumor or tumor progression to the metastatic state
EP0923335B1 (en) * 1996-07-08 2006-08-23 Animas Corporation Implantable sensor and system for in vivo measurement and control of fluid constituent levels
US5922714A (en) * 1996-12-26 1999-07-13 Bionumerik Pharmaceuticals, Inc. Combination of MDAM and 5-fluoropyrimidines for treating cancer
US5798526A (en) * 1997-01-24 1998-08-25 Infrasoft International Llc Calibration system for spectrographic analyzing instruments
EP1011426A1 (en) * 1997-02-26 2000-06-28 Diasense, Inc. Individual calibration of blood glucose for supporting noninvasive self-monitoring blood glucose
ES2281143T3 (en) * 1997-11-12 2007-09-16 Lightouch Medical, Inc. METHOD FOR THE NON-INVASIVE MEASUREMENT OF AN ANALYTE.
US6119026A (en) * 1997-12-04 2000-09-12 Hewlett-Packard Company Radiation apparatus and method for analysis of analytes in sample
US6864978B1 (en) * 1999-07-22 2005-03-08 Sensys Medical, Inc. Method of characterizing spectrometer instruments and providing calibration models to compensate for instrument variation
US6216022B1 (en) * 2000-06-22 2001-04-10 Biosafe Laboratories, Inc. Implantable optical measurement device and method for using same
US6549861B1 (en) * 2000-08-10 2003-04-15 Euro-Celtique, S.A. Automated system and method for spectroscopic analysis
US6615151B1 (en) * 2000-08-28 2003-09-02 Cme Telemetrix Inc. Method for creating spectral instrument variation tolerance in calibration algorithms

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6601582B2 (en) * 1997-06-10 2003-08-05 Smithkline Beecham Corporation Dispenser with doses' counter
WO2003063699A1 (en) * 2002-01-25 2003-08-07 Sensys Medical, Inc. Indirect measurement of tissue analytes through tissue properties
WO2004000113A1 (en) * 2002-06-25 2003-12-31 Sensys Medical, Inc. Targeted interference subtraction applied to near-infrared measurement of analytes
US20050240090A1 (en) * 2003-03-07 2005-10-27 Ruchti Timothy L Method of processing noninvasive spectra

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102519944B (en) * 2006-03-10 2015-04-01 康宁股份有限公司 Optimized method for lid biosensor resonance detection
CN108226064A (en) * 2016-12-15 2018-06-29 株式会社堀场制作所 Analytical equipment, computer readable storage medium and analysis method

Also Published As

Publication number Publication date
US20060017923A1 (en) 2006-01-26
WO2006017700A3 (en) 2007-06-07
US7436511B2 (en) 2008-10-14

Similar Documents

Publication Publication Date Title
US20060017923A1 (en) Analyte filter method and apparatus
AU738441B2 (en) Method and apparatus for generating basis sets for use in spectroscopic analysis
US7620674B2 (en) Method and apparatus for enhanced estimation of an analyte property through multiple region transformation
Lasch Spectral pre-processing for biomedical vibrational spectroscopy and microspectroscopic imaging
US7640140B2 (en) Method of processing noninvasive spectra
EP0877926B1 (en) Apparatus for multi-spectral analysis in noninvasive nir spectroscopy
AU716192B2 (en) Method and apparatus for multi-spectral analysis in noninvasive infrared spectroscopy
US6441388B1 (en) Methods and apparatus for spectroscopic calibration model transfer
US5750994A (en) Positive correlation filter systems and methods of use thereof
US7098037B2 (en) Accommodating subject and instrument variations in spectroscopic determinations
WO1997028437A9 (en) Method and apparatus for multi-spectral analysis in noninvasive infrared spectroscopy
WO2009038908A1 (en) Calibrated analyte concentration measurements in mixtures
US20210270742A1 (en) Peak-preserving and enhancing baseline correction methods for raman spectroscopy
Gributs et al. Parsimonious calibration models for near-infrared spectroscopy using wavelets and scaling functions
Tan et al. Improvement of a standard-free method for near-infrared calibration transfer
Chen et al. Removal of major interference sources in aqueous near-infrared spectroscopy techniques
Patchava et al. Savitzky-Golay coupled with digital bandpass filtering as a pre-processing technique in the quantitative analysis of glucose from near infrared spectra
CN116509386A (en) Noninvasive in-vivo blood glucose detection system and method based on Raman spectrum
AU713502C (en) Method and apparatus for multi-spectral analysis in noninvasive NIR spectroscopy
Barman Investigation of constrained regularization for the development of a robust and clinically accurate multivariate calibration procedure

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A2

Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BW BY BZ CA CH CN CO CR CU CZ DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KM KP KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NA NG NI NO NZ OM PG PH PL PT RO SC SD SE SG SK SL SM SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW

AL Designated countries for regional patents

Kind code of ref document: A2

Designated state(s): BW GH GM KE LS MW MZ NA SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IS IT LT LU LV MC NL PL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
122 Ep: pct application non-entry in european phase
DPE2 Request for preliminary examination filed before expiration of 19th month from priority date (pct application filed from 20040101)