CA2464634A1

CA2464634A1 - Pap estimator

Info

Publication number: CA2464634A1
Application number: CA002464634A
Authority: CA
Inventors: Boris Popov; Victor F. Lanzo; Rajeev Agarwal
Original assignee: Andromed Inc
Current assignee: Andromed Inc
Priority date: 2004-04-16
Filing date: 2004-04-16
Publication date: 2005-10-16
Also published as: EP1744666A1; US7909772B2; US20120071767A1; JP2007532207A; CA2569730A1; WO2005099562B1; WO2005099562A1; US20080091115A1

Description

TITLE OF THE INVENTION
PAP ESTIMATOR
BACKGROUND OF THE INVENTION
Pulmonary artery hypertension is caused by several heart or pulmonary diseases including dysfunction of prosthetic or native heart valves, left ventricular dysfunction, congenital abnormalities of the heart and great vessels, chronic obstructive pulmonary disease, and adult respiratory distress syndrome. Pulmonary artery hypertension is a serious cardiovascular dysfunction that is difficult to assess non-invasively. In patients requiring continuous monitoring . or with those suspected with Pulmonary artery hypertension, the Pulmonary Artery Pressure (PAP) is usually measured using a pulmonary arterial catheter. This is an invasive surgical procedure that is associated with significant morbidity and mortality. A pulmonary arterial catheter can be left in place for a few days to allow continuous monitoring of PAP of patients in the critical care unit. However, due to the potential risk to the patients, it is not recommended for repeated measurements that are sometimes necessary (more than 2 days continuous measurements, once per week, once per month or every 6 months depending on the evolution of the disease). Nonetheless, regular evaluation of the PAP is very important to identify pulmonary hypertension and subsequently to follow the evolution of the disease and to assess the efficacy of the treatment. Consequently, non-invasive methods are necessary to allow frequent and accurate measurement of PAP.
Doppler echocardiography is one non-invasive method that can be used to estimate the systolic PAP. if tricuspid regurgitation i s detected, it is possible to estimate the systolic pressure gradient across the tricuspid valve using continuous-wave Doppler. The right ventricular systolic pressure can be calculated by adding the systolic tricuspid valve gradient to the estimated right atrial pressure. The right ventricular systolic pressure can be considered equivalent to the systolic PAP when the systolic pressure gradient across the pulmonary valve is negligible. This noninvasive method can provide a high degree of correlation (0.89 < r < 0.97) and a standard error estimate (SEE) varying from 7 to 12 mmHg with pulmonary artery catheterization (for systolic PAP range: 20-160 mmHg). However, the estimation of PAP by Doppler echocardiography has several important limitations. First, PAP cannot be estimated by Doppler in approximately 50% of patients with normal PAP, 10 to 20 % of patients with elevated PAP, and 34 to 76% of patients with chronic obstructive pulmonary disease because of the absence of tricuspid regurgitation, a weak Doppler signal, or a poor signal-to-noise ratio. To improve the feasibility of the method in some of these patients, it is necessary to use contrast agent enhancement. Second, Doppler echocardiography tends to overestimate PAP in patients with normal PAP and significantly underestimate the PAP in patients with severe pulmonary arterial hypertension.
Another non-invasive PAP estimation method is based on the spectral properties of the P2 component of the second heart sound. The basic ~rincipie supporting this approach is based on Laplace's Law and assumes that the tension of the pulmonary wall is proportional to the corresponding intra-arterial blood pressure. Similar to a stretched drumhead, it is expected that the resonant frequency of the blood column into the pulmonary artery is proportional to the tension in the arterial wall, and thus to the arterial pressure.
Spectral methods have shown a high correlation between spectral features of P2 and the systolic PAP measured by PA catheterization. The two features were the dominant frequency and the quality factor of the spectrum of P2. In patients with prosthetic heart valves it has been demonstrated that systolic PAP
estimated by Doppler echocardiography can be predicted (r = 0.84, SEE = ~ 5 mmHg, P < 0.0001 ) from the spectra! features extracted from S2.

Another approach for estimating PAP has been proposed in US Patent No.
6,368,28381 using advanced signal processing techniques on S2 heart sounds and incorporated herein by reference. The splitting tirne interval (SI) between the aortic (A2) and the pulmonary (P2) components of S2 is measured using a computer-assisted (but not fully automated) spectral de-chirping method and is normalized for heart rate - Normalized SI (NSI). Here, the idea is that the NSI is proportional to PAP.
BRIEF DESCRIPTION OF THE DRAWINGS
In the appended drawings:
Figure 1 discloses an illustrative embodiment of a PAP estimator according to an illustrative embodiment of the present invention;
Figure 2 discloses typical signals detected using an ECG and a pair of biological sound monitors (BSM1 and BSM 2) )according to an illustrative embodiment of the present invention; and Figures 3A and 3B disclose a flow chart of the A2, P2 and NSI detection portion of the PAP estimator according to an illustrative embodiment of the present invention.
DETAILED DESCRIPTION OF THE ILLUSTRATIVE EMBODIMENTS
Referring now to Figure 1, an illustrative embodiment of a PAP Estimator, generally referred to using the reference numeral 10, will now be described.
Illustratively, two identical biologics! sound sensors 12, for example those described in US Patent No. 6,661,161 are provided for, although in a given application a different number of sensors may be preferable. The sensors 12 are placed at two different locations on the patient 14. One sensor 12~ is positioned at the apex of heart, where the A2 component of the S2 sound is likely at its maximal in intensity and P2 component is minimal. A second sensor 122 is placed to maximize the P2 component intensity (between the 3'd and 4th left intercostal space). The best sensor locations are obtained by experimenting with different positions while observing S2 sound signals, so as to achieve the maxima! signal intensity.
The sensors 12 are attached via appropriate leads as in 16 to a data acquisition system 18 comprised of an analog to digital converter 20 and personal computer 22. Data collected by the sensors 12 is digitised by the analog to digital converter 20, illustratively using a sampling rate of 2k~lz with 12 bits of resolution. Additionally, EEG signals are also collected via a series of electrodes 24, leads 26 and a second analog to digital converter 28. Similar to the acoustic data collected by the biological sound sensors 12, data collected by the EEG electrodes 24 is digitised by the analog to digifial converter 28, illustratively using a sampling rate of 2kHz with 12 bits of resolution. As will be seen below, the electrocardiogram is used as the reference signal to frame the second heart sound (S2).
Referring now to Figure 2, an Electrocardiogram (ECG) reading is displayed along side readings from first and second biological sound sensors.
Automatic A2 and P2 Detection The electrocardiogram is used to provide the reference signal to frame the second heart sound (S2). The beat signal in the description below means the part of acoustic signal between two consecutive QRS complexes on the ECG.
For each beat signal the first heart signal (S1 ) is detected and removed. The remaining sounds, including the second heart sounds and possibly murmurs and the like, are used as input.

Referring now to the flow charts of Figures 3A and 3B in addition to Figure 1, an illustrative embodiment of an approach for estimating the NSI will now be described. That algorithm supports input signals from the two sensors 12, each of them comprised of signals of heart sounds in the frequency range 30-200 Hz, although this range could be wider without any changes in the approach. If that range is narrower, however, the algorithm shoulld be adapted to those limitations.
Sounds related to heart beats are collected at 100 via a sensor 12 and divided into three sub channels 102, 104 and 106 (or frequency bands). These bands are: Low Frequency (LF, 30-50 Hz), Medium Frequency (MF, 50 -150 Hz), and High Frequency (HF, 120-200 Hz).
Each sub-channel is relayed to a "Process Channel" block as in 108, 1082, and 1083, (these will be described separately hereinbelow). The process channel block can be based on a variety of algorithms including a Chirplet algorithm, Non-linear Energy Operator (NLEO) algorithm, or any other suitable algorithm capable of extracting and discriminating A2 and P2 components from second heart sound S2.
Of note is that the present illustrative embodiment applies the NLEO
algorithm.
The output values of A2 and P2 from the process channel blocks as in 108, 1082, and 1083 are analysed. If both components A2, I'2 are clearly detectable in at least one of the sub channels, these are the values for A2, P2. If both components are not clearly detectable then the outputs of the process channel blocks as in 108, 1082, and 1083 are compared sub-channel by sub-channel with the output of the process channel blocks for other sensors (not shown) of the same sub channels at blocks 110, 112, and 114. In the case at hand, there are illustratively two sensors (the second sensor not shown) the outputs of the process blocks of which are thus compared pair wise.

_6_ Illustratively, the comparison is carried out on each frequency band according to the following set of rules, although it should be understood that this is an example and not intended to be limiting:
~ If the output of 108 for both sensors reveals A2 and P2 components and the positions of A2 and P2 are the same, then these positions provide the values of A2 and P2;
~ If one of the outputs of 108 for both sensors reveals A2 and P2 components, but the other does not, then the positions of these A2 and P2 provide the values of A2 and P2;
~ If the output of 108 for both sensors reveals only one A2 or one P2 component then, as it is unknown whether the component is A2 or P2, then the value of A2 is the position of the first component and the value of P2 the position of the second component.
~ If the output of 108 for one of the sensors rE;veals both A2 and P2 components while the output of 108 for the other sensor reveals only one (A2 or P2) component, then the reading: for both sensors are combined (superimposed).
o If the result reveals only two component, (A2 and P2) then the positions of these A2 and P2 provide the values of A2 and P2;
o If the result still reveals three components (where one or two of the results are A2 andlor P2 and the remainder the result of biological noise), then the readings are combined (superimposed) and the two components with the greateat FWE are selected as A2 and P2, the positions of these A2 and P2 provide the values of A2 and P2.
~ If the output of 108 for both sensors reveals A2 and P2 components but the positions of A2 and P2 are different, then:
o If the Splitting Interval (SI) of both sensor; is less than 10ms then the value of A2 is the position of A2 and the value of P2 is the position of P2 as determined via one of the sensors;
o If at least one of the SI from first or second sensor is greater than 10ms, all components (A2 and P2) within 10 ms are merged.
~ If only one component results, then the value of both A2 and P2 is the position this one component and resulting SI
is equal to zero;
~ If two components result, then the value of A2 is the position of the first component and the value of P2 the position of the second component;
~ If three components result, then the values of A2 and P2 are the positions of the two components with the greatest FWE; and ~ If four components result, then the values of A2 and P2 are the positions of A2 and P2 from the sensor where the amplitude of components FWE is greater than that of the other sensor.
The NSI for each sub-channel, including combined channels, is also calculated.
The A2 and P2 components in the LF, MF, and HF sub-channels have small variation in positioning because of different frequency content. As a result, at block 116, heuristic rules are used to correct those deviations and produce A2 and P2 single values from the combination of A2 and P2 from all sub-channels (LF, MF, HF) as well as any combined values which may have been generated.
An illustrative example of the heuristic rules applied at block 116 is as follows:
~ If no values for both A2 and P2 (from which the NSI is calculated) are available in the MF and HF sub-channels and i:he NSI of the LF channel > 120msec, then discard the NSI of the LF channel;
~ If no values for both A2 and P2 are available in the LF and HF sub-channels and the NSI of the LF channel > 1.4 * NSi of the HF channel, then discard the NSI of the LF channel;
~ If no values for both A2 and P2 are available in the LF and MF sub channels, and the NSI of the LF channel > 1.4 * the NSI of the MF
channel, then the NSI of the LF channel = 1.4 * the NSI of the MF
channel; and ~ If no values for both A2 and P2 are available in the MF and HF sub-channels, and the NSI of the MF channel < 1.4 * the NSI of the HF
channel, then the NSI of the HF channel = (111.4) * the NSI of the MF
channel.
Referring now to Figure 3B, the normalised values of A2, P2 and NSI for the current beat are calculated at blocks 118, 120 and 122 and stored at blocks 124, 126 and 128. Illustratively, values of NA2, NP2 and NSI calculated for beats during the previous minute are retained.
At the same time consistency of solution and signal-to-noise ratio (SNR) for each sub-channel is estimated and stored in separated lists. In this regard, for each sub-range the SNR is estimated. Consistency indicates the percentage of beats not rejected due to high noise. Illustratively, in order to determine the SNR, the S2 sound is first detected as well as the precise position of start and end of S2. The signal component (S) is calculated as the energy between the start and end of S2, divided by the duration of S2 (in cosec). The noise component (N) is calculated as the energy within 50 cosec segment before S2 start added to the energy within 50 cosec segment after S2 end divided by 100 cosec. The resulting signal-to-noise ratio is calculated as SNR = SIN.
After all beats within the time averaging interval {in the case at hand illustratively 1 minute) have been processed in the above manner, a series of values of A2, P2, NSI are ready for statistical validation. At a first step of that validation the laws of distribution of A2 and P2 are estimated and threshold value T calculated using the bias criterion. Typically between 50-200 beats are present during a one minute sampling interval. Histograms are used in order to _g_ provide an estimation of the distribution laws. The distribution law of NSI is used for additional control of the T value in the case of rnulti-peak distribution or A2 or P2.
At block 130, any values of A2 which are greater than T and values of P2 less than T are discarded from the stored values. The NSI values are then recalculated at block 132 using only those A2 and P2 values which still have pairs.
At blocks 134, 136 and 138 the central peaks on A2, P2 and NSI histograms are estimated using two-iteration algorithm. Then at block 140 the value NSI' _ P2-A2 is calculated.
At block 142, NSI' is compared with the peak value of NSI calculated at block 138. If the difference between NSI and NSI' is less than 0.01 (or 1 % in terms of average beat duration), the mean value of NSI and NSI' is produced as the final output value for NSI. If a difference between NSI and NSI' is greater than 0.01, the value of NSI, NSI' with the higher consistency value, as previously calculated at blocks 144, 146 is produced as the final output value.
Referring back to Figure 3A, as stated hereinabove, the process channel block 108 can be based on a variety of algorithms including a Chirplet algorithm, NLEO algorithm, or any other suitable algorithm capable of extracting and discriminating A2 and P2 components from second heart sound S2.
Illustratively, the NLEO algorithm is described and comprises the following processing steps. Referring to block 1082, The Signal to Noise Ratio (SNR) is determined at block 148.
At decision block 150, if the SNR is below a predeternnined value (illustratively 1.5), the current beat is discarded and no further processing steps carried out, giving rise to an empty set of A2, P2 at block 152 (in effect, these values are flagged to indicate that this particular beat is noisy and should not be used for any combinations. Alternatively, if the SNR is above a predetermined value the NLEO function is calculated at block 154 using the current beat's signal.
In this regard, the NLEO function (also known as the Frequency Weighted Energy (FWE) criterion) or any other individual implementation of general family of Autocorrelators can be used.
NLEO is a simple manipulation of digital signal described in the general case by:
'I'~n~=x(n-l)'x~wm)-x~wP)'x~ny) foYl+m=p+q (1) One of NLEO's properties is the ability to compactly describe the notion of frequency-weighted energy (FWE), which is different from the mean-square energy as it reflects both the amplitude as well as the frequency content of a signal. In the case of a pure tone the output can be described as:
F (A*sin (w*n)J = A2*sin (w*(I p+q-s)/2) * sin (w*(q-s-I+p)l2) (2) For the special case where l ~ m and p ~ q, given an input of additive white Gaussian noise (AWGN) the expected value of NLEO output is zero. Thus it has the ability to suppress noise. If we consider the case of amplitude modulated short duration sinusoidal burst in the presence of random noise and structured sinusoidal interference (as in the case of the aortic and the pulmonary components of the S2 sound in the midst of noise), it is anticipated that the NLEO output will enhance FWE of each of these components while suppressing AWGN interference and provide a constant baseline for sinusoidal interference. The time-varying nature of amplitude (Gaussian) and chirping of the dominant rhythm will modulate the NLEO output and produce a detectable burst corresponding to each component in contrast to background clutter. It will then be possible to apply detection strategies on the NLEO output with S2 sound input.

Illustratively, NLEO with parameters I = 2, m = 3, p = 1, q = 4 was applied.
For those parameters the output is:
FjA*sin(w*T)J = A2*sin(wT) * sin(2wT) = 0.5*(cos(wT) - cos(3wT)) (3) Once the NLEO function is calculated, at block 15fi the highest peak (maximum of NLEO output for given beat signal) is determined and those peaks having values of less than 0.05 of highest peak value are removed. In this regard, 0.05 provides good results, although other values may also provide adequate results. If more than two peaks remain, the A2 and P2 candidates are identified at block 158. If only one peak is detected, then this is passed to the output and determined as A2 or P2 according to the procedure described hereinabove at paragraph 18.
Finally, at block 160 the values of A2 and P2 are validated using list of heuristic rules. An illustrative example of such rules are:
~ if time interval between A2 and P2 on NLEO is greater than 100 msec, the component with lower FWE is invalid;
~ if time interval between A2 and P2 on NLEO is less than 10 msec, check for those components on original S2 signal. If there is no separated components, the component with lower FWE is invalid; and ~ if there is difference more then 10 times befinreen FWE of A2 and P2 components, the component with lower FWE is invalidated.
Although the present invention has been described hereinabove by way of an illustrative embodiment thereof, this embodiment can be modified at will, within the scope of the present invention, without departing from the spirit and nature of the subject of the present invention.

Claims

1. A method for estimating the PAP of a patient, the method comprising the steps of:
collecting a series of second heart sounds (S2) from the patient from at least two sensors;
for each of said sensors and said second heart sounds:
filtering said second heart sound into a plurality of frequency bands; and in each of said frequency bands, deriving the aortic component (A2) and pulmonary component (P2) from said filtered heart sound;
wherein if A2 and P2 are not readily detectable in at least one of said frequency bands, A2 and P2 derived for each of said frequency bands are compared with those of the others of said frequency bands and other sensors using an autocorrelator in order to generate a combined value of A2 and P2; and determining a normalised splitting interval (NSI) for A2 and P2 of said series of second heart sounds.

2. The method of claim 1, wherein said autocorrelator is NLEO.