WO2012069887A1 - Method and device for electroencephalogram based assessment of anesthetic state during anesthesia or sedation - Google Patents

Abstract

A method and device to monitor the anesthetic state of animal patients undergoing general anesthesia are provided. Electroencephalogram derived parameters are commonly used tools for anesthetic depth monitoring in human patients. However, there are no objective measures of anesthetic depth monitoring available for veterinary use. Furthermore, the parameters available for humans have showed inadequacy in animals. A robust invention which is adaptable to the species anesthetized and to the anesthetic drugs used is presented. This invention transforms (5, 6, 7, 8) the animal's EEG (4) into an easy to interpret parameter (10) that reflects the anesthetic depth. This parameter undergoes continuous optimization according to the anesthesiologists' input (9) regarding the animal anesthetic state, species and anesthetic protocol. This function performs fine tuning resulting in a parameter which correlates with anesthetic state, optimized for each species and anesthetic protocol, helping the anesthesiologist to adjust the drug doses and to automatically control the administration of anesthesia.

Description

D E S C R I P T I O N

"METHOD AND DEVICE FOR ELECTROENCEPHALOGRAM BASED ASSESSMENT OF ANESTHETIC STATE DURING ANESTHESIA OR

SEDATION"

Technical Field

The present invention relates to a method and apparatus for monitoring a condition of an animal under anesthesia or sedation, in particular, information acquired from different sources from the animal is used.

Summary

The present invention describes a method for assessment of a condition of a patient during anesthesia or sedation comprising the steps of:

acquiring (1) a signal representing EEG activity;

deriving from said signal a first parameter value of linear analysis (5);

deriving from said signal a second parameter value of non-linear analysis (6);

deriving from said signal a third parameter value of suppression quantification (7);

applying (8) a predetermined mathematical index for probability of animal comfort, in which index at least said parameters are variables;

calculating successively changing values of said probability index; and

indicating (10) said successive index values. In a preferred embodiment said predetermined mathematical index for probability of animal comfort is obtained by combining (8) said parameters (5, 6, 7) with weights retrieved from a database (11), at least said weights are according to animal species and anaesthetic protocol.

In a preferred embodiment said predetermined mathematical index for probability of animal comfort is obtained by combining (8) the three sub-parameters L - linear (5), NL - non-linear (6) and SQ - suppression quantification (7) by [a*L+ (100 -a) *NL ] * (l-SQ/100) , where a is the weighting factor for L and NL, all variables in base 100.

A preferred embodiment additionally comprises the step of pre-processing (2) by filtering, amplifying and converting to digital the signal representing EEG activity after acquiring ( 1 ) .

A preferred embodiment further comprises previously detecting and removing artifacts by causing a new recording of an EEG fragment, before the calculation of said parameters .

A preferred embodiment further comprises periodically or intermittently storing the recorded EEG (4) and the anesthesiologist input (9) of anesthetic depth in the database (11) .

A preferred embodiment further comprises offline optimization, at least according to animal species and anaesthetic protocol, comprising the steps of:

calculating, in the combinatory unit (8), the possible combinations of weighted fractions for the parameters (5, 6, 7), stored in the database (10), extracted from previously recorded EEG (4) together with combinations for the factors used in the suppression quantification (7) ;

calculating the performance of those combinations, in the comparator unit (12), by correlating it to the anesthesiologist's input (9) regarding animal anesthetic depth observed previously stored in the database (11) ;

storing in the database (11) the weighting and suppression quantification (7) factors that produced the combination of the parameters that showed the best performance . a preferred embodiment the calculation of:

the correlation of the performance of the combinations with the anesthesiologist's input (9);

the factors that showed the best performance;

is obtained by prediction probability through a measure of association, by an artificial neural network or by a fuzzy inference system.

A preferred embodiment further comprises in addition to animal species and anaesthetic protocol, the further calculation of said predetermined mathematical index for probability of animal comfort with the weights of said parameters (5, 6, 7) according to breed, sex, weight and physical status of the animal or surgical procedure.

In a preferred embodiment the deriving from said signal of a first parameter value of linear analysis (5), is obtained through spectral edge frequency, by the frequency value which is below a predetermined value of spectral power. In a preferred embodiment the deriving from said signal of a second parameter value of non-linear analysis (6), is obtained through standard ordinal pattern analysis by permutation entropy calculation with the standard Shannon uncertainty formula.

In a preferred embodiment the deriving from said signal of a third parameter value of suppression quantification (7), is obtained through a burst suppression ratio, calculated by the percentage of periods in which EEG is isoelectric, its amplitude being within predetermined limits for a minimum predetermined period of time.

A preferred embodiment further comprises the steps of:

verifying if for a predetermined period of time, there are not suppression periods;

if it is the case, alerting and requiring the user to manually confirm the anesthetic depth of the patient and to input that information;

then, if the patient is confirmed in deep anaesthesia, adapting the suppression quantification limit by decreasing the upper and lower limits until suppression can be identified, and storing the new limit information .

The present invention also describes a device for the assessment of a condition of a patient during anesthesia or sedation comprising:

signal acquisition unit (1) for a signal representing

EEG activity;

linear analysis unit (5), connected to said signal acquisition unit (1); non-linear analysis unit (6), connected to said signal acquisition unit (1);

suppression quantification analysis unit (7), connected to said signal acquisition unit (1);

combinatory unit (8) connected to the output of said analysis units (5, 6, 7);

output interface (10), connected to said combinatory unit ( 8 ) .

A preferred embodiment further comprises a database (11), containing records of weights for said parameters (5, 6, 7) at least according to animal species and anaesthetic protocol .

In a preferred embodiment said combinatory unit (8) operates by combining (8) the three sub-parameters L linear (5), NL - non-linear (6) and SQ - suppression quantification (7) by [ a*L+ (100 -a) *NL ] * ( l-SQ/100 ) , where a is the weighting factor for L and NL, in base 100.

A preferred embodiment additionally comprises a signal pre^¬ processing unit (2) for filtering, amplifying and converting to digital the signal representing EEG activity at the output of the signal acquisition unit (1) .

A preferred embodiment further comprises an artifact detection unit (3) comprising a sub-unit (3.2) able to trigger a new recording of an EEG fragment should an artifact be present.

A preferred embodiment further comprises:

an anesthesiologist input unit (9) a comparator unit (12), able to calculate the performance of combinations of weighted fractions for the parameters (5, 6, 7), stored in the database (11) by correlating it to the anesthesiologist's input (9) regarding animal anesthetic depth observed previously stored in the database (11) .

In a preferred embodiment the database unit (11) further comprises records, in addition to animal species and anaesthetic protocol, the further calculation of said parameters according to breed, sex, weight and physical status of the animal or surgical procedure.

In a preferred embodiment the linear analysis unit (5) is a spectral edge frequency calculator, of the frequency value which is below a predetermined value of spectral power; the non-linear analysis unit (6) is a calculator of standard ordinal pattern analysis by permutation entropy calculation with the standard Shannon uncertainty formula; the suppression quantification unit (7) is a calculator of classic burst suppression ratio, by the percentage of periods in which EEG is isoelectric, its amplitude being below predetermined limits for a minimum predetermined period of time.

A preferred embodiment comprises a data processing system comprising code means adapted to carry out each of the steps of the methods above.

A preferred embodiment comprises a computer program comprising code means adapted to perform the steps of any of the methods above when said program is run on a data processing system. A preferred embodiment comprises a computer program as above embodied on a computer-readable medium.

Background Art

The present invention relates to the field of veterinary anesthesia, in particular to the intraoperative and postoperative monitoring of animals' anesthetic depth.

In the present, no direct measures are available for monitoring the anesthetic depth in veterinary patients.

Veterinary anesthetists adjust anesthetic drug dosing by assessing changes in clinical signs of the patient during anesthesia, such as blood pressure, heart rate, pupil dilation, sweating, lacrimation, movement, etc. However, these signs are not reliable indicators of the anesthetic depth, as they are reflex in nature and do not mirror cortical depression caused by anesthetic drugs. Furthermore, the evaluation of these signs is dependent on the skill and experience of the veterinarian, on the anesthetized species and on the anesthetic protocol used. General anesthesia in animals represents thus a serious problem due to the difficulties in accessing the anesthetic depth. This incapacity to know if the animal is too deep, too light or in the correct anesthetic depth results in inefficient drug adjustment which leads to over or under dosage, two potential dangerous and deleterious situations which can result in death or disease by one side or in awareness and pain sensation during surgery, by the other. Because general anesthetics' target effect site is the central nervous system, the electrical signal recorded from the brain (electroencephalogram - EEG) has been used as a measure of anesthetics effect, reflecting the depth of anesthesia in human patients.

The interpretation of the unprocessed raw EEG is very complex and time consuming, which lead to the development of several EEG-based parameters that are extracted from the EEG and represented as a simple number that varies in a scale from 0 to 99 or 100, indicating if the patient is too deeply anesthetized (when the value is below 40) or too superficial (when the value is above 60) .

Between those parameters, the Bispectral Index described in documents U.S. Pat. Nrs 4.907.597, 5010.891, 5.320.109 e 5.458.117 is the most widely known. It is employed as a routine tool in operating theatres all over the world. This invention's output is a weighted sum of different parameters that are mainly derived using spectral and higher order spectral analysis and that are based in a data base from hundreds EEG's recorded from anesthetized human patients .

There are several studies exploring the potential of the Bispectral Index for anesthetic depth monitoring in animals including rats, rabbits, cats, dogs, pigs, horses. The application of this monitor in animals was in general, not successful .

Other EEG-based anesthetic depth parameters that exist for human patients as commercial monitors include the Cerebral State Index, based in a combination of sub-parameters extracted from the EEG and combined by fuzzy-logic, commercialized by Danmeter ®. This monitor has been tested in dogs but its capacity to detect the effects of general anesthesia was not satisfactory.

There are other parameters, such as the Spectral Entropy, described in document U.S. Pat. No. 6.801.803 and commercialized by GE Healthcare® and the Index of Consciousness commercialized by Aircraft Medical Barcelona® which are based in different mathematical methods than Bispectral Index and Cerebral State Index. However, all methods are based in the recording of the spontaneous electroencephalogram in the anesthetized patient, and in its transformation into a simple and objective number which varies between 0 and 100.

In animals, the existence of a wide range of species and wide variability between individuals inside the same species turns it difficult to develop a unique fixed parameter for such a diverse use. Nevertheless, the existence of a tool for animal anesthetic depth monitoring is a urgent need for veterinarians.

The existence of this tool is crucial to decrease the mortality rate due to anesthesia in animals which is still too high reaching 2% in dogs and cats and 8% in horses, while in humans it is about 0.01%.

The general failure found in human EEG-derived parameters for monitoring anesthesia in animals, can be explained partly by their calibration to the characteristics of human EEC

The fact that the development of the Bispectral Index was also based on analysis of a database of EEG in humans may explain the lack of precision of this equipment when used in animals.

However, the incapacity of the methods to detect anesthetic effects on animals EEG may be overcome by the use of a robust parameter which has small sensitivity to external artifacts such as movement or the electrical activity of the muscles - electromyographic activity. This is especially important when applying the method to a wide range of different head anatomies, with different head muscle layers thickness. In this context, the use of nonlinear analysis techniques based in order pattern analysis showed promising results for the analysis of animals EEC The Permutation entropy has recently started to be applied for EEG analysis. It is an ordinal pattern analysis parameter that can reveal the nonlinear characteristics of EEG and has high robustness to artifacts, an essential feature for its practical application .

The power spectral analysis of the EEG is the traditional and most simple method to extract its information. It reveals frequency changes caused by anesthetics, which are linear and well known in animals therefore providing a very easy to interpret response to anesthetics.

At superficial anesthetic planes, the electroencephalogram shows a characteristic shift to high amplitude and low frequency waves, which increases with increasing dose and is reflected in a decrease in power spectral parameters and permutation entropy extracted from the EEG. However, when anesthesia is deeper a specific EEG pattern appears - the burst suppression pattern. This is characterized by high frequency and high amplitude periods alternated with periods of EEG silence. The high frequency components cause a paradoxical increase in power spectral parameters and permutation entropy, turning them inefficient in reflecting anesthetic depth during deeper anesthesia.

Disclosure of the Invention

The object of the present invention is, namely, the use of three combined parameters which are capable of detecting different EEG characteristics during anesthesia (a parameter for non-linear analysis, a parameter for power spectral analysis and a parameter for quantification of the EEG suppression periods) and express the EEG signal in a easy to understand parameter. This method undergoes continuous training that allows an increase in devices performances for the most commonly anesthetized species and most used anesthetic protocols.

This method could be incorporated in devices that record the animal's EEG and used to aid the veterinarian in monitoring the anesthetic depth of different species, both during surgical procedures and in the intensive care unit, where patients are often kept sedated under mechanical ventilation during long periods. The use of this tool in these situations will help the practitioners to adjust the sedative and anesthetic drugs more accurately, avoiding side effects and deleterious consequences of over and under dosage. This method may also be used for neurophysiology diagnostics, as it displays the EEG providing the means to monitor cases of epilepsy, cerebral ischemia and to help diagnosing brain death.

It is important to stress that all the phases are performed non-invasively on the body and the invention defines neither a diagnostic nor a clinical picture of pathology.

The present invention provides a device and method for animal anesthetic depth monitoring based on the spontaneous EEG recorded from the animal's head.

More specifically, the invention provides a device and a method that receives the animal EEG, shows the unprocessed EEG waveform and transforms it into a parameter which can be used by the anesthesiologist to access the animal anesthetic depth and adjust the anesthetic dosage comprising : Signal acquisition unit (1) in which the EEG is recorded from the animal's head.

Signal pre-processing unit (2) where the EEG signal is filtered, amplified and converted to digital information .

Artifact detection Unit: Artifacts are then identified and EEG fragments with artifacts are removed from the analysis ( 3 ) .

Signal processing Units (4-12) : The pre-processed and artifact-free EEG (4) undergoes transformation into three sub-parameters: linear parameter (L) , non-linear parameter (NL) and suppression quantification (SQ) (5,6,7) which are combined in a combinatory unit (8) previously trained to produce the final parameter that best reflects the anesthetic depth of that animal species and anesthetic protocol which is displayed in the anesthesiologist interface (10).

Training is performed in the training unit (12) and obtained by:

5.1. Initial training using an initial database (11) with EEG (4) recorded preferably from several species and anesthetic protocols and the respective anesthesiologist input (9).

5.2. Continuous training using the EEG (4) and the anesthesiologyst ' s input (9) preferably recorded in each use of the device in veterinary practices which are continuously stored on the updated database (11) . The anesthesiologist input (9) includes preferably:

6.1. Initial anesthesiologist's input, regarding the animal species, breed, weight, sex, general physical status, surgical procedure and anesthetic protocol to be used. The animal species and anesthetic protocol are, in particular, used to organize recorded data on the database.

6.2. Intermittent input, during the whole anesthetic procedure, regarding the anesthetic depth observed by the anesthesiologist through the observations of clinical signs during the real time use of the method. This input is incorporated in the method, preferably in the form of a numerical scale used in the training.

The final parameter is then calculated as follows by combining the three sub-parameters (5,6,7) extracted from the EEG (4) according to the function:

[a*L+ (100-a) *NL] * ( 1-SQ (t , v, -v) ) in which a is the weighting factor for L and NL and varies between 0 and 100; t is the minimum time to considered the EEG as suppressed (for the SQ quantification) and v and -v are the voltage limits between which the EEG is considered as suppressed (for the SQ quantification) .

In the training phase, the calculation of the Final Parameter is performed in different manners by changing the combinatory factors a, t, and v and -v , which leads to the generation of multiple Final Parameters. These multiple Final Parameters are then compared regarding their ability to reflect the anesthesiologyst input (9) stored in the database. This can be performed using, for example, a method of prediction probability comparison, applied to all the data contained in the database, in which an offline analysis is performed that determines the Final Parameter that works best, by correlating the values of each different Final Parameter with the correspondent clinical scale of anesthesia stored. The ability of the Final Parameters to track changes in the clinical scale of anesthesia is calculated by applying the prediction probability measure, which is a measure of correlation that was specifically developed to study the performance of depth of anesthesia indexes. Pk ranges from 0 to 1. A value of Pk = 0.5 means that the indicator correctly predicts the anesthetic depths only 50% of the time, i.e., no better than a 50:50 chance. A value of Pk = 1 means that the indicator predicts the anesthetic depths correctly 100% of the time.

An artificial neural network system or a fuzzy inference system can be preferably used for the training phase, performed in the training unit (12) after original data acquisition and to increase predictability.

8. As the database increases, new elements are included in the training set, namely new EEG data (4) and anesthesiologysts input (9) resulting in fine-tuning of the final parameter throughout use.

9. Increasing use of the invention and data acquisition does allow the method to be more specific taking into account other input variables recorded by the anesthesiologist, as the breed, the sex and the weight, physical status or surgical procedure. Thus, the final parameter may easily be adapted to different clinical situations .

The invention includes a system for remote analysis of the data recorded and supervision of the training performed in a way that only valid data are kept on the database and used for training. Updates are then to be sent to each device in use in veterinary practices.

Follows an example of correlation values obtained between the anesthetic depth evaluated using a numerical scale of anesthesia, in this case from 1 (awake) to 5 (deep anesthesia) and : each of the sub-parameteres (linear - L, non-linear - NL and suppression quantification - SQ) and the final parameter calculated by combination of the three subparameters in six rabbits. While the linear and non^¬ linear sub-parameters do not show any significant correlation with the clinically observed anesthetic scale, their combination on a final parameter shows a correlation coefficient (Pearson r) of -0.90. By combining different sub-parameters derived from the EEG it is thus possible to obtain a parameter that decreases with increasing doses of anesthesia, reflecting the deepening in anesthetic depth thus achieving one of the main goals of the present invention .

Description of the Figures

Figure 1. Example of fragments of electroencephalogram (EEG) recorded in rabbits anesthetized with propofol. All fragments were recorded at a sampling rate of 256 samples/second and 8 seconds of recording are shown. A - EEG recorded in an awake rabbit; B - EEG recorded in a rabbit under superficial anesthesia with a plasma propofol concentration of 13 pg/ml; C - EEG recorded in a rabbit under deep anesthesia with a plasma concentration of propofol of 30 pg/ml.

Figure 2. Power spectral analysis of EEG fragments shown in Figure 1. The Fast-fourier transform was used to compute the power spectrum. A - Power spectral representation of the EEG recorded in an awake rabbit; B - Power spectral representation of the EEG recorded in a rabbit under superficial anesthesia with a plasma propofol concentration of 13 pg/ml; C - Power spectral representation of the EEG recorded in a a rabbit under deep anesthesia with a plasma concentration of propofol of 30 pg/ml. The dashed line represents the spectral edge frequency 95% obtained from each EEG fragment.

Figure 3. Method for the derivation of the non-linear component based in permutation entropy. A - Example of the fragmentation of the EEG in motifs. A small fragment of EEG (with 0.07 seconds duration) is shown. It is separated into motifs (vectors of three data points) . Each motif is underlined with a grey dotted line.

Bl to B6 - The method further identifies each motif as belonging to one of the six possible types (vectors in squares: Bl to B6) according to their shape.

CI to C6 - It counts the number of motifs from the real EEG that belongs to each of the six categories, to obtain the probability of occurrence of each motif in the signal.

D - The method finally calculates the permutation entropy of the resultant normalized probability distribution of the motifs, using the standard Shannon entropy, resulting in a final single value for a determined EEG fragment.

Figure 4. Burst Suppression Pattern during propofol anesthesia in a rabbit. Eight seconds EEG are shown at 256 samples/second. The high amplitude waves ("bursts") are alternated with periods of low amplitude and low frequency ("suppression") .

Figure 5. Example of the application of the suppression quantification method in three fragments of EEG: A recorded in an awake rabbit - no suppression is found. B - recorded in a rabbit under superficial anesthesia - no suppression is found; C - recorded in a rabbit under deep anesthesia - a period of suppression is found between around 0-1 and 3-4 seconds of recording. Between 5-6 seconds a period of EEG shows the burst suppression pattern, however it is not quantified because the time unit was not reached.

Figure 6. Architecture of the device for the EEG pre^¬ processing .

1- Signal acquisition Unit

2- Signal pre-processing Unit

3.1- Unit that in the case of no artifacts present on the signal outputs said artifact-free EEG

3.2 - Unit that in the case of detection of signal artifacts produces feedback message, displays a signal of incorrect data acquisition, which indicates for new recording of EEG fragment before re-sending to the Processing Unit

4 - Pre-processed artifact-free EEG storage unit Figure 7. Architecture of the device for its real-time use during anesthesia

4- Pre-processed artifact-free EEG storage unit

5- Linear analysis (L) unit

6- Non-linear analysis (NL) unit

7- Suppression quantification (SQ) unit

8 - Combinatory Unit for 5, 6 and 7.

9 - Anesthesiologysts input Unit. Comprises initial information regarding animal species, weight, sex, physical status and anesthetic protocol used as well as intermittent input in the form of a scale, for example from 1 to 5 regarding the clinical anesthetic depth as evaluated by the anesthesiologyst. In the real-time use of the device this input is important to determine the combinatory function used to produce the final parameter, as this can be different between species and anesthetic protocols.

10 - Final output interface showed to the anesthesiologist.

Figure 8. Architecture of the device for the training phase .

4- Pre-processed artifact-free EEG storage unit

5- Linear analysis (L) unit

6- Non-linear analysis (NL) unit

7- Suppression quantification (SQ) unit

9 - Anesthesiologysts input Unit. Comprises initial information regarding animal species, weight, sex, physical status and anesthetic protocol used as well as intermittent input in the form of a scale, for example from 1 to 5 regarding the clinical anesthetic depth as evaluated by the anesthesiologyst.

11 - Database Unit that stores the EEG (4) and anesthesiologysts ' input (9) recorded in every anesthesia. 12 - Training Unit which incorporates the database (11) and trains the device to calculate and combine the sub- parameters (5,6 and 7) in proportions that better reflect the anesthesiologysts input (9) for a specific species and anesthetic protocol. Training can be performed using a method of prediction probability comparison, artificial neural network or fuzzy inference systems.

Detailed Description and Preferred Embodiments

The present invention provides a method for animal anesthetic depth monitoring based on the spontaneous EEG recorded from the animal's head.

More specifically, the invention provides a method that receives the animal EEG and the anesthesiologist's input regarding the animal and the anesthetic protocol characteristics and performs the optimization of an EEG- derived parameter which can be further used to access the animal anesthetic depth and adjust the anesthetic dosage. The device and method preferably comprises:

1- Signal acquisition Unit (1)

a. The EEG is recorded from the animal's head using surface electrodes with a sampling frequency of namely 1024 Hz (1024 points per second) .

2- Signal pre-processing Unit (2)

a. The EEG is optionally first amplified with namely a gain xlOOO. (2)

b. A band-pass filter (namely 0.1-300 Hz) and a notch filter (namely 50 Hz) are optionally applied (2)

c. The signal is converted to digital by an Analogue-to-Digital converter (2) i. Converts the amplitude of the potential recorded at each point in time to a number that is stored.

3- Artifact detection unit: the signal is sent to the microprocessor unit which analyzes possible artifacts. (3) . If the fragment of EEG analyzed has significant external artifacts, the system sends a message for the anesthesiologist in the form of a sound alarm, so that he readjusts the electrodes in the patient's head until a fragment with sufficient quality can be recorded (3.1). If no artifacts are detected (3.2), the fragment is sent to the processing unit (4) .

4- Signal processing Unit (4-12) .

Receives the pre-processed artifact-free EEG fragments (4) and performs real time analysis transforming the EEG into a parameter while the animal is under anesthesia.

This is performed by extracting three sub-parameters (Linear analysis - L (5), non-linear analysis - NL (6) and suppression quantification - SQ (7)) from the EEG and combining them in weighted fractions to produce a final parameter in a combinatory unit (8) . The weighting fractions of L and NL are given by the trained database (11) which is under continuous training in the training unit ( 12 ) .

Three EEG samples are shown in figure 1 for the awake state (1-A), for superficial anesthesia (1-B) and for deep anesthesia (1- C) .

Each of these sub-methods is capable of processing detect different characteristics of the signal during anesthesia and are extracted as follows; Linear analysis - L (5)

The linear analysis performed by the method is based on the parameter spectral edge frequency (SEF) 95%. This parameter is acquired after power spectral analysis of the signal by the Fast fourier transform (FFT) . FFT is a function expressed in terms of sinusoidal basis functions, allowing the analysis of EEG frequency components. The application of the Fourier transform results in EEG spectral power in which the power of the frequency bands that comprise the signal is calculated. The parameter SEF is calculated on the basis of this power spectrum, consisting of the frequency below which lies 95% of spectral power. Other values are possible, namely 70%, 90% or 99%. Figure 2 shows the power spectral analysis of the EEG fragments shown previously in Figure 1. Figure 2-A shows the power spectrum of the EEG fragment recorded in an awake rabbit (shown in Figure 1-A) ; Figure 2-B shows the power spectrum of the EEG fragment recorded in a rabbit under superficial anesthesia, with a plasma concentration of the anesthetic propofol of 13 pg/ml, as shown in figure 1-B. Figure 2-C shows the power spectrum of the EEG fragment presented in Figure 1-C, which was recorded in a rabbit under deep anesthesia, with a propofol plasma concentration of 30 pg/ml. The SEF of the power spectrum is shown for each referred anesthetic depth. It is a parameter commonly used to characterize the EEG during anesthesia and has the great advantage of being very quick and simple calculation.

Non linear analysis - NL(6)

Although the electroencephalogram signal shows characteristic changes during anesthesia regarding the frequency components that can be detected by the linear analysis parameter, this type of signal also has nonlinear behavior, which demand the application of methods capable of detecting such changes, to avoid losing important information from the EEC

The nonlinear analysis included in this method is based on the method of ordinal pattern analysis: permutation entropy which is calculated according to the following steps:

1- Fragments the continuous EEG signal into a sequence of motifs with length = 3. Figure 3-A shows an example of this fragmentation performed in a small fragment of EEG (with 0.07 seconds duration) . As shown, this fragment is separated into motifs (vectors of three data points). Each motif is underlined with a grey dotted line.

2 - Identifies each motif as belonging to one of the six possible types (Figure 3- from Bl to B6) according to their shape .

3 - Counts the number of motifs from the real EEG of each that belongs to each of the six categories, to obtain the probability of occurrence of each motif in the signal (Figure 3 - from CI to C6) .

4 - Calculates the permutation entropy of the resultant normalized probability distribution of the motifs, using the standard Shannon uncertainty formula (Figure 3-D) :

For the shown examples in Figure 1, the final results for this non-linear analysis for the EEG recorded in the awake state, superficial anesthesia and deep anesthesia were, for 1-A (awake) 0.76, for 1-B (superficial) 0.64 and for 1-C (deep) 0.69.

Suppression quantification analysis - SQ (7)

At superficial anesthetic planes, the electroencephalogram shows a characteristic shift to high amplitude and low frequency waves, which increases with increasing dose and is reflected in a decrease in power spectral parameters and permutation entropy extracted from the EEG. However, when anesthesia is deeper a specific EEG pattern appears - the burst suppression pattern. Figure 4 shows the burst suppression pattern in fragment of EEG of 8 seconds. This is characterized by high frequency and high amplitude periods alternated with periods of EEG silence. The high frequency components cause a paradoxical increase in power spectral parameters and permutation entropy, turning them inefficient in reflecting anesthetic depth during deeper anesthesia. This is shown in Figures 2 (from B to C) and in the result from the non-linear parameter calculation from figure 3-B (as the value increases from superficial (final result = 0.64) to deep (final result = 0.69)).

For the recognition of these patterns in the EEG of animals, it is proposed to include a component in the method that detects fragments of low signal amplitude.

The present invention includes a component for suppression quantification based in the classic burst suppression ratio defined for human patients.

This parameter is equal to the percentage of periods in which EEG is isoelectric.

This is expressed in a percentage which is further incorporated in the combination with the other subparameters by multiplication of (l-SQ/100). Two factors are important when determining the SQ: the voltage limits in which the EEG is considered isoelectric (v and -v) and the minimum time during which the EEG is isoelectric to consider it as suppressed (t) .

The EEG is normally considered to be isoelectric when its amplitude is between -5 and 5 microvolts for a minimum time period of 400 miliseconds. However, this value may be altered by the recording conditions, such as the electrode placement techniques, recording conditions, and animal head anatomy. This can be adapted, without prejudice to the present invention, as the skilled person sees fit, or the evolution of the field and the specific animal characteristics and anesthetic protocols used so demand. This adaptation is part of the training performed in the training unit (12) .

As shown in Figure 5, during the awake state (5-A) and superficial (5-B) anesthesia no suppression is observed. During deep anesthesia (5-C) the amplitude limits can detect suppression.

This method also includes a system for manual confirmation of suppression. The inclusion of this component was motivated by the fact that in humans, there are some reports of failure of commercial monitors in detecting electroencephalographic silence, which resulted in a wrong conclusion from the anesthesiologist, increasing the dose of anesthetic to the patient. This is a very dangerous situation, as it can result in overdose and must be avoided. In this invention, depending on the anesthetic protocol used and animal species anesthetized an alarm will be displayed when there are increases in the linear and non-linear sub-parameters, but the method or device is not able to find suppression periods. The equipment gives an alarm sound and displays a message on the user interface so that the anesthesiogist manually confirms the anesthetic depth of the patient and inserts that information in the device. Two possible situations may be found:

1- The anesthetic depth is insufficient and the patient needs more anesthetic.

2- The anesthetic depth is considered deep, but the

signal amplitude is different from usual, preventing the suppression quantification function to work properly .

If the patient is in deep anesthesia, the method or device then adapts the suppression quantification limit by sliding the upper and lower limits until suppression can be identified and stores that limits information. This adaptation is accompanied by the anesthesiologysts input (9) to confirm the possibility of BS patterns occurrence at that anesthetic depth. This information is then considered in the training unit (12) to improve subsequent uses of the invention .

Combinatory Unit (8)

The three sub-parameters L (5), NL (6) and SQ (7) are then combined by the combinatory unit according to the function:

Final parameter = [ a*L+ (100 -a) *NL ]*(1-SQ (t,v, -v)/100);

Where a is the weighting factor for L and NL which varies from 0 to 100 and t and v are respectively the time and voltage limits for the classification of the EEG as suppressed .

The values for a, t and v and -v are given initially by an initial database and throughout use they are regularly updated after training with the growing database. They may thus be adapted to different species and anesthetic protocols, to produce the final parameter that better reflects the anesthetic depth in different conditions.

Training Unit (12)

Training is obtained by:

- Initial training using an initial database (11) with EEG (4) recorded preferably from several species and anesthetic protocols and the respective anesthesiologists' input (9) . The method has an initial trained database incorporated that determines the initial values for a, t, v and -v depending on the species anesthetized and the anesthetic protocol used .

- Continuous training using the EEG (4) and the anesthesiologyst ' s input preferably recorded (9) in each use of the device in veterinary practices.

The anesthesiologist input (9) preferably includes:

- Initial anesthesiologist's input, regarding the animal species, breed, weight, sex, general physical status, surgical procedure and anesthetic protocol to be used. The animal species and anesthetic protocol are in particular used to organize recorded data on the database .

- Intermittent input, during the whole anesthetic procedure, regarding the anesthetic depth observed by the anesthesiologist through the observations of clinical signs during the real time use of the method. This input is incorporated in the database, for example in the form of a numerical scale from 1 (awake) to 5 (deep anesthesia) and is used in the training .

The three sub-parameters (5,6,7) extracted from the EEG (4) are combined according to the function mentioned above. Training consists in adapting the calculation of the final parameter, by changing the combinatory factors a,t and v, producing multiple Final Parameters and comparing their capacity to reflect the clinical depth of anesthesia. The Final Parameter that shows the best capacity to reflect depth of anesthesia can thus be elected and the values attributed to a, t and v used in subsequent devices' utilization. To study and compare the performance of the multiple Final Parameters obtained a method of prediction probability analysis can be used.

Prediction probability (Pk) has been established in anesthesia as a statistical method to assess the capability of a parameter to discern different levels of anesthesia. The aim of the Pk analysis is to quantify the association between the (clinically) observed anesthetic level and the parameter values; in this case of different versions of a parameter based on different weights of sub-parameters L and NL . It is a type of nonparametric correlation known as a measure of association which is suited to ordinal variables and can accommodate variable scales having any degree of coarseness or fitness.

Pk is a variant of Kim's d_y *symbol* x measure of association. Kim's d_y *symbol* x is defined for ordinal variables x and y in terms of the types of pairs of data points just described. Let P_c, Pc and P_tx be the respective probabilities that two data points drawn at random, independently and with replacement, from the population are a concordance, a discordance, or an x-only tie.

The only other possibility is that the two data points are tied in observed depth y; therefore, the sum of P_c, P_d, and P sub tx is the probability that the two data points have distinct values of observed anesthetic depth, that is, that they are not tied in y.

The Pk can thus be represented by: c * r¾ 4- r'tx

Pk range is from 0 to 1. When the probabilities of discordance and indicator-only tie are both zero, Pk = 1. When the probability of discordance equals that of concordance, Pk = 0.5. A value of Pk equal or lower than 0.5, means that discordances are more likely than concordances .

Specifically, given two randomly selected data points with distinct observed anesthetic depths, Pk is the probability that the indicator values of the data points predict correctly which of the data points is the lighter (or deeper) . A value of Pk = 0.5 means that the indicator correctly predicts the anesthetic depths only 50% of the time, i.e., no better than a 50:50 chance. A value of Pk = 1 means that the indicator predicts the anesthetic depths correctly 100% of the time.

Alternatively, an artificial neural network system or a fuzzy inference system can preferably be used for this training phase, performed in the training unit (12) .

10. As the database grows, new elements are included in the training set, namely new EEG data (4) and anesthesiologysts input (9) resulting in fine-tuning of the final parameter throughout use.

11. Increasing use of the invention and data acquisition does allow the method to be more specific taking into account other input variables recorded by the anesthesiologist, as the breed, the sex and the weight, physical status or surgical procedure. Thus, the final parameter may easily be adapted to different situations. The invention includes a system for remote control of the data recorded and supervision of the training performed that only valid data are kept on the database. Updates can then be sent to each device in use by regular software upgrades .

As an example, if a dog is anesthetized with the drug propofol, the method can use the combination of parameters L and NL previously optimized through training with data recorded in dogs under propofol anesthesia and time (t) and voltage limits (v,-v) for the SQ calculation previously used in dogs under propofol anesthesia to derive the final parameter. During the animal's anesthesia the anesthesiologysts input is stored in the database to be used in the further training for this species and anesthetic resulting in the fine-tuning of these values for subsequent applications.

The innovations of this method are namely:

1- It is the first method for animal EEG processing during anesthesia.

2- It applies three sub-parameters and combines them into a single and easy to understand parameter in real time.

3- The information displayed may allow the anesthesiologist to adjust the drugs administration .

4- The extracted and displayed parameter may be continuously optimized to species and anesthetic protocol . 5- Consequently, it can be used in a wide variety of species, first for the method optimization, and then as a tool for monitoring the animal anesthetic depth .

6- It stores important data regarding the anesthetic procedure in an automatic way, which is not possible in the present in veterinary anesthesia where anesthetic data recording is performed manually .

7- It may also be optimized for more specific features such as animal sex, age, breed, physical condition, surgical procedures.

By displaying the raw EEG it can also be used in for neurophysiology diagnostic aids of epilepsy and brain death .

The following claims set out particular embodiments of the invention .

Claims

C L A I M S

1. Method for assessment of a condition of an animal patient during anesthesia or sedation characterized in that it comprises the steps of:

- acquiring (1) a signal representing EEG activity;

- deriving from said signal a first parameter value of linear analysis (5);

- deriving from said signal a second parameter value of non-linear analysis (6);

- deriving from said signal a third parameter value of suppression quantification (7);

- applying (8) a predetermined mathematical index for probability of animal comfort, in which index at least said parameters are variables;

- calculating successively changing values of said probability index; and

- indicating (10) said successive index values.

2. Method according to claim 1 characterized in that said predetermined mathematical index for probability of animal comfort is obtained by combining (8) said parameters (5, 6, 7) with weights retrieved from a database (11), at least said weights are according to animal species and anaesthetic protocol.

3. Method according to the previous claims characterized in that said predetermined mathematical index for probability of animal comfort is obtained by combining (8) the three sub-parameters L - linear (5), NL - non^¬ linear (6) and SQ - suppression quantification (7) by [a*L+ (100 -a) *NL ] * (l-SQ/100) , where a is the weighting factor for L and NL, all variables in base 100.

4. Method according to the previous claims characterized in that it aditionally comprises the step of pre^¬ processing (2) by filtering, amplifying and converting to digital the signal representing EEG activity after acquiring ( 1 ) .

5. Method according to the previous claims characterized in that it further comprises previously detecting and removing artifacts by causing a new recording of an EEG fragment, before the calculation of said parameters.

6. Method according to the previous claims characterized in that it further comprises periodically or intermittently storing the recorded EEG (4) and the anesthesiologist input (9) of anesthetic depth in the database (11) .

7. Method according to the previous claim characterized in that it further comprises offline optimization, at least according to animal species and anaesthetic protocol, comprising the steps of:

- calculating, in the combinatory unit (8), the possible combinations of weighted fractions for the parameters (5, 6, 7), stored in the database (10), extracted from previously recorded EEG (4) together with combinations for the factors used in the suppression quantification (7);

- calculating the performance of those combinations, in the comparator unit (12), by correlating it to the anesthesiologist's input (9) regarding animal anesthetic depth observed previously stored in the database (10) ;

- storing in the database (11) the weighting and suppression quantification (7) factors that produced the combination of the parameters that showed the best performance.

8. Method according to the previous claim characterized in that the calculation of:

- the correlation of the combinations with the anesthesiologist's input (9);

- the factors that showed the best performance;

is obtained by prediction probability through a measure of association, by an artificial neural network or by a fuzzy inference system.

9. Method according to the previous claims characterized in that it further comprises in addition to animal species and anaesthetic protocol, the further calculation of said predetermined mathematical index for probability of animal comfort with the weights of said parameters (5, 6, 7) according to breed, sex, weight and physical status of the animal or surgical procedure .

10. Method according to the previous claims characterized in that the deriving from said signal of a first parameter value of linear analysis (5), is obtained through spectral edge frequency, by the frequency value which is below a predetermined value of spectral power.

11. Method according to the previous claims characterized in that the deriving from said signal of a second parameter value of non-linear analysis (6), is obtained through standard ordinal pattern analysis by permutation entropy calculation with the standard Shannon uncertainty formula.

12. Method according to the previous claims characterized in that the deriving from said signal of a third parameter value of suppression quantification (7), is obtained through a burst suppression ratio, calculated by the percentage of periods in which EEG is isoelectric, its amplitude being within predetermined limits for a minimum predetermined period of time.

13. Method according to the previous claims characterized in that it further comprises the steps of:

- verifying if for a predetermined period of time, there are not suppression periods;

- if it is the case, alerting and requiring the user to manually confirm the anesthetic depth of the patient and to input that information;

- then, if the patient is confirmed in deep anaesthesia, adapting the suppression quantification limit by decreasing the upper and lower limits until suppression can be identified, and storing the new limit information.

14. Device for assessment of a condition of an animal patient during anesthesia or sedation characterized in that it comprises:

- signal acquisition unit (1) for a signal representing EEG activity;

- linear analysis unit (5), connected to said signal acquisition unit (1); - non-linear analysis unit (6), connected to said signal acquisition unit (1);

- suppression quantification analysis unit (7), connected to said signal acquisition unit (1);

- combinatory unit (8) connected to the output of said analysis units (5, 6, 7) configured to calculate a predetermined mathematical index for probability of animal comfort, in which index at least said analysis unit (5, 6, 7) outputs are variables;

- output interface (10), connected to said combinatory unit ( 8 ) .

15. Device according to claim 14 characterized in that it further comprises a database (11), containing records of weights for said parameters (5, 6, 7) at least according to animal species and anaesthetic protocol.

16. Device according to claims 14 - 15 characterized in that said combinatory unit (8) operates by combining (8) the three sub-parameters L - linear (5), NL - non^¬ linear (6) and SQ - suppression quantification (7) by [a*L+ (100 -a) *NL ] * (l-SQ/100) , where a is the weighting factor for L and NL, in base 100.

17. Device according to claims 14 - 16 characterized in that it aditionally comprises a signal pre-processing unit (2) for filtering, amplifying and converting to digital the signal representing EEG activity at the output of the signal acquisition unit (1) .

18. Device according to claims 14 - 17 characterized in that it further comprises an artifact detection unit (3) comprising a sub-unit (3.2) able to trigger a new recording of an EEG fragment should an artifact be present .

19. Device according to claims 14 - 18 characterized in that it further comprises:

- an anesthesiologist input unit (9)

- a comparator unit (12), able to calculate the performance of combinations of weighted fractions for the parameters (5, 6, 7), stored in the database (10) by correlating it to the anesthesiologist's input (9) regarding animal anesthetic depth observed previously stored in the database (11) .

20. Device according to claims 14 - 19 characterized in that the database unit (11) further comprises records, in addition to animal species and anaesthetic protocol, the further calculation of said parameters according to breed, sex, weight and physical status of the animal or surgical procedure.

21. Device according to claims 14 - 20 characterized in that the linear analysis unit (5) is a spectral edge frequency calculator, of the frequency value which is below a predetermined value of spectral power; the non^¬ linear analysis unit (6) is a calculator of standard ordinal pattern analysis by permutation entropy calculation with the standard Shannon uncertainty formula; the suppression quantification unit (7) is a calculator of classic burst suppression ratio, by the percentage of periods in which EEG is isoelectric, its amplitude being below predetermined limits for a minimum predetermined period of time.

22. A data processing system comprising code means adapted to carry out each of the steps of claims 1 - 13.

23. A computer program comprising code means adapted to perform the steps of any claim 1 - 13 when said program is run on a data processing system.

24. The computer program as claimed in the previous claim embodied on a computer-readable medium.