WAVELET BASED FEATURE EXTRACTION AND DIMENSION REDUCTION FOR THE CLASSIFICATION OF HUMAN CARDIAC ELECTROGRAM DEPOLARIZATION WAVEFORMS

• US 20080109041A1
• Filed: 11/01/2007
• Published: 05/08/2008
• Est. Priority Date: 11/08/2006
• Status: Active Grant

• Current Assignee: Biotronik SE & Company KG
• Original Assignee: Biotronik CRM Patent AG (Biotronik SE & Company KG)
• Inventors: de Voir, Christopher
• Granted Patent: US 7,751,873 B2
• Time in Patent Office: Days
• US Class Current: 607/7
• CPC Class Codes: A61B 5/316 Modalities, i.e. specific diagnostic methods
• CPC Class Codes: A61B 5/33 specially adapted for cooperation with other devices
• CPC Class Codes: A61B 5/347 Detecting the frequency distribution of signals
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/7203 for noise prevention, reduction or removal
• CPC Class Codes: A61B 5/7232 involving compression of the physiological signal, e.g. to extend the signal recording period
• CPC Class Codes: A61B 5/7239 using differentiation including higher order derivatives
• CPC Class Codes: A61B 5/7257 using Fourier transforms
• CPC Class Codes: A61B 5/726 using Wavelet transforms
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: A61N 1/3621 for treating or preventing abnormally high heart rate
• CPC Class Codes: A61N 1/37 Monitoring; Protecting
• CPC Class Codes: A61N 1/3956 Implantable devices for applying electric shocks to the heart, e.g. for cardioversion
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

A depolarization waveform classifier based on the Modified lifting line wavelet Transform is described. Overcomes problems in existing rate-based event classifiers. A task for pacemaker/defibrillators is the accurate identification of rhythm categories so correct electrotherapy can be administered. Because some rhythms cause rapid dangerous drop in cardiac output, it'"'"'s desirable to categorize depolarization waveforms on a beat-to-beat basis to accomplish rhythm classification as rapidly as possible. Although rate based methods of event categorization have served well in implanted devices, these methods suffer in sensitivity and specificity when atrial/ventricular rates are similar. Human experts differentiate rhythms by morphological features of strip chart electrocardiograms. The wavelet transform approximates human expert analysis function because it correlates distinct morphological features at multiple scales. The accuracy of implanted rhythm determination can then be improved by using human-appreciable time domain features enhanced by time scale decomposition of depolarization waveforms.

Claims

1. A monitoring device for monitoring and evaluation of a physiological signal including a cardiac electrogram to detect physiological events and/or categorize physiological events and discriminate them from each other by way of comparison to prototype events comprising:

a monitoring device comprising a signal preprocessing stage having an input for a digitized physiological signal being a digital representation of a time course of a physiological signal;

said signal preprocessing stage comprising an event-detector that is adapted to determine a fiducial point for at least one segment of said physiological signal by detecting reoccurring characteristic features within said physiological signal including R-waves or p-waves;

said monitoring device further comprising a wavelet coefficient generation stage that is adapted to perform a modified lifting line wavelet transformation on a segment of said physiological signal relative a selected fiducial point in a signal segment to thus generate wavelet transformation coefficients as features of said digitized physiological signal;

a feature evaluation stage that is adapted to perform a reduction of features of said digitized physiological signal by selecting those wavelet transformation coefficients that have been determined to be most characteristic of prototype events from each other by means of a receiver operation characteristics curve analysis;

a feature threshold memory comprising a threshold value for each selected feature of a prototype event, said threshold value being determined by said receiver operation characteristics curve analysis; and

,

a comparator stage connected to said feature threshold memory and being adapted to compare said selected features of an actual digitized physiological signal to corresponding threshold values for a prototype event and to assign said actual digitized physiological signal to a prototype event category if each or a majority of said selected features of said actual digitized physiological signal is within a threshold defined by a corresponding one of said threshold values.

2. The monitoring device according to claim 1, wherein said wavelet coefficient generation stage is adapted to perform said modified lifting line wavelet transformation based on a high pass detail filter and a corresponding low path smoothing filter to thus obtain wavelet detail and scale coefficients, respectively.

3. The monitoring device according to claim 2, wherein said wavelet coefficient generation stage is adapted to perform said modified lifting line wavelet transformation recursively thus obtain further wavelet detail coefficients for a next scale based on smoothed values determined while determining detail coefficients for a previous scale.

4. The monitoring device according to claim 1, wherein said signal preprocessing stage has at least one further input for additional input signals, such as an accelerometer value or cardiac impedance values.

5. The monitoring device according to claim 1, wherein said signal preprocessing stage is adapted to generate a feature vector from the wavelet coefficients produced by said wavelet coefficient generation stage and said further input values for further processing by said feature evaluation stage.

6. The monitoring device according to claim 1, wherein said feature evaluation stage is adapted to perform feature evaluation in relation to wavelet coefficients produced from a signal segment centered on a fiducial point of each signal segment to be evaluated.

7. The monitoring device according to claim 1, wherein said monitoring device comprises a threshold determination stage comprising a receiver operation characteristics analyzer that is adapted to process features including wavelet coefficients of a regular event and another specific prototype event that are time aligned with each other based on said fiducial point to thus determine the most characteristic features and to further determine and store a threshold value for each of said most characteristic features in said threshold memory.

8. The monitoring device according to claim 1, wherein said threshold memory comprises more than one set of threshold values wherein each of said threshold values corresponds to a particular prototype event and wherein the comparator stage is adapted to compare said selected features of an actual digitized physiological signal to corresponding threshold values for each prototype event separately.

9. The monitoring device according to claim 1, wherein said comparator stage comprises a kernel discriminant analyzer adapted to perform a kernel discriminant analysis on said selected features of an actual digitized physiological signal to assign said selected features of an actual digitized physiological signal to one of a plurality of prototype events used to determine a kernel discriminant for discrimination between the prototype events.

16. A heart stimulator comprising a signal sensing stage and a monitoring device according to claim 1 wherein said input of said preprocessing stage is connected to said sensing stage.

17. The heart stimulator according to claim 16, wherein said sensing stage comprises a band path filter having a pass band between 5 Hz and 100 Hz.

18. The heart stimulator according to claim 16, wherein said sensing stage is adapted to be connected or is connected to a bipolar electrode lead.

19. The heart stimulator according to claim 16, wherein said heart stimulator is an implantable pacemaker and/or an implantable cardioverter/defibrillator.

10. A heart therapy system comprising an implantable heart stimulator and an external device comprising:

an implantable heart stimulator comprising a signal sensing stage;

a signal preprocessing stage having an input for a digitized physiological signal being a digital representation of a time course of a physiological signal, said input being connected to said signal sensing stage;

said signal preprocessing stage comprising an event-detector that is adapted to determine a fiducial point for at least one segment of said physiological signal by detecting reoccurring characteristic features within said physiological signal such as R-waves or p-waves;

an external device comprising

a wavelet coefficient generation stage that is adapted to perform a modified lifting line wavelet transformation to thus generate wavelet transformation coefficients as features of said digitized physiological signal;

a feature evaluation stage that is adapted to perform a reduction of features of said digitized physiological signal by selecting those wavelet transformation coefficients that have been determined to be most characteristic for discriminating a prototype event from a regular event by means of a receiver operation characteristics curve analysis;

a feature threshold memory comprising a threshold value for each selected feature of a proto-type event, said threshold value being determined by means of said receiver operation characteristics curve analysis; and

,

a comparator stage connected to said threshold memory and being adapted to compare said selected features of an actual digitized physiological signal to corresponding threshold values for a prototype event and to assign said actual digitized physiological signal to the prototype event if each or a majority of said selected features of said actual digitized physiological signal is within the threshold defined by a corresponding one of said stored threshold values.

11. The heart therapy system according to claim 10 wherein said sensing stage comprises a band path filter having a pass band between 5 Hz and 100 Hz.

12. The heart therapy system according to claim 10 wherein said sensing stage is adapted to be connected or is connected to a bipolar electrode lead.

13. A heart therapy system comprising an implantable heart stimulator and an external device comprising:

an implantable heart stimulator comprises a signal sensing stage;

a signal preprocessing stage having an input for a digitized physiological signal being a digital representation of a time course of a physiological signal, said input being connected to said sensing stage;

said signal preprocessing stage comprising an event-detector that is adapted to determine a fiducial point for at least one segment of said physiological signal by detecting reoccurring characteristic features within said physiological signal such as R-waves or p-waves;

an event detecting stage comprising

a wavelet coefficient generation stage that is adapted to perform a modified lifting line wavelet transformation to thus generate wavelet transformation coefficients as features of said digitized physiological signal;

a feature evaluation stage that is adapted to perform a reduction of features of said digitized physiological signal by selecting those wavelet transformation coefficients that have been determined to be most characteristic ones for discriminating a prototype event from a regular event by means of a receiver operation characteristics curve analysis;

a feature threshold memory comprising a threshold value for each selected feature of a prototype event, said threshold value being determined by means of said receiver operation characteristics curve analysis;

a comparator stage connected to said threshold memory and being adapted to compare said selected features of an actual digitized physiological signal to corresponding threshold values for a prototype event and to assign said actual digitized physiological signal to the prototype event if each or a majority of said selected features of said actual digitized physiological signal is within the threshold defined by a corresponding one of said stored threshold values;

an external device comprising a threshold determination stage comprising a receiver operation characteristics analyzer that is adapted to process features including wavelet coefficients of a regular event and a prototype event that are time aligned with each other based on said fiducial point to thus determine the most characteristic features and to further determine a threshold value for each of said most characteristic features to be in said threshold memory of said implantable heart stimulator.

14. The heart therapy system according to claim 13, wherein said sensing stage comprises a band path filter having a pass band between 5 Hz and 100 Hz.

15. The heart therapy system according to claim 13, wherein said sensing stage is adapted to be connected or is connected to a bipolar electrode lead.

20. A method for monitoring and evaluation of a periodic or quasi-periodic signal such as a cardiac electrogram to detect physiological events and/or discriminate physiological events from each other by way of comparison to prototype events comprising:

signal preprocessing to determine a fiducial point for at least one segment of said physiological signal by detecting reoccurring characteristic features within said physiological signal such as R-waves or p-waves;

wavelet coefficient generating via a modified lifting line wavelet transformation to thus generate wavelet transformation coefficients;

feature evaluating including reduction of features of said digitized physiological signal based on said wavelet transformation coefficients by means of receiver operation characteristics curve analysis to determine a reduced set of most characteristic wavelet coefficients; and

,

feature comparing to compare said reduced set of wavelet coefficients to one or more stored reduced sets of wavelet coefficients, each stored reduced set of wavelet coefficients corresponds to a prototype event.