BIOMEDICAL SIGNAL MORPHOLOGY ANALYSIS METHOD

• US 20100049069A1
• Filed: 11/30/2007
• Published: 02/25/2010
• Est. Priority Date: 12/01/2006
• Status: Active Grant

• Current Assignee: Obs Medical Limited
• Original Assignee: Oxford Biosignals (Us) Incorporated
• Inventors: Patterson, Alan, Strachan, Guy, Tarassenko, Lionel
• Granted Patent: US 8,290,575 B2
• Time in Patent Office: Days
• US Class Current: 600/512
• CPC Class Codes: A61B 5/1455 using optical sensors, e.g. spectral photometrical oximeters
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/364 Detecting abnormal ECG interval, e.g. extrasystoles, ectopic heartbeats
• CPC Class Codes: A61B 5/369 Electroencephalography [EEG] devices for psychotechnics A61B5/16
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A way of quantifying the shape of an ECG waveform is disclosed by detecting the JT segment using two Hidden Markov Models and calculating the analytic signal of the JT segment. Parameters calculated from the analytic signal are used as shape descriptors for the JT segment. The shape descriptors may be displayed in a dimensionality-reduced mapping. Templates representing characteristic shapes can be produced by finding cluster centres in the shape descriptor space, and the novelty of new waveforms can be quantified by comparing the position in shape descriptor space of new shape descriptors to a predefined normal training set or to previously encountered waveforms. Novel shape descriptors can be used to retrieve the corresponding waveforms, and templates of such novel shapes can be created by averaging such waveforms, using dynamic time warping to allow for variations in heart rate. The templates can be manually segmented and the manual segmentation propagated back into other waveforms having similar shape descriptors.

Claims

1. A method of analysing a biomedical signal having a repetitive signal feature, comprising the steps of segmenting each of a plurality of said repetitive signal features in the biomedical signal, analysing one or more of the segments to find the values of a plurality of parameters describing the shape of said one or more of the segments recording the values, and tracking changes in the said values through the biomedical signal.

2. A method according to claim 1 wherein the biomedical signal is an electrocardiogram and the repetitive signal feature is the heartbeat

3. A method according to claim 2 wherein the step of analysing one or more of the segments comprises analysing the T-wave of the heartbeat, being the segment from the J point to the T_OffSet.

4. A method according to claim 3 wherein the J point is detected by use of a Hidden Markov Model comprising a state model for the waveform after the J point.

5. A method according to claim 2 wherein the T_Offset is detected by use of a Hidden Markov Model comprising state models for the T wave and the waveform after T_Offset.

6. A method according to claim 5 wherein T_OffSet is detected by the steps of:

maintaining from beat to beat through the electrocardiogram an estimate of the QT interval based on a relationship between the QT interval and the heart rate, estimating the position of T_OffSet from the estimated QT interval, and applying the Hidden Markov Model comprising state models for the T wave and the waveform after T_Offset to a portion of the electrocardiogram defined with respect to the estimated T_offset position.

7. A method according to claim 6 wherein a Kalman filter is used to maintain said estimate of the QT interval.

8. A method according to claim 1 wherein said plurality of parameters describing the shape of a segment constitute components of a shape descriptor vector for that segment.

9. A method according to claim 1 further comprising the step of defining normality for said shape descriptor vector, and calculating for each of said plurality of repetitive signal features the difference between its shape descriptor vector and normality.

10. A method according to claim 9 wherein normality is defined by shape descriptor vectors based on a plurality of repetitive signal features from training set.

11. A method according to claim 10 wherein the step of calculating for each of said plurality of repetitive signal features the difference between its shape descriptor vector and normality comprises calculating the distance in shape descriptor vector space between its shape descriptor vector and the nearest shape descriptor vector based on the training set.

12. A method according to claim 10 wherein the step of calculating for each of said plurality of repetitive signal features the difference between its shape descriptor vector and normality comprises calculating the density in the shape descriptor vector space of points derived from the training set, and calculating the probability of normality of the shape descriptor vector for each of said plurality of repetitive signal features.

13. A method according to claim 9, further comprising the step of displaying a morphology index value based on said difference.

14. A method according to claim 13 wherein changes in the morphology index value with time through the biomedical signal are displayed

15. A method according to claim 8 thereon further comprising the step of applying a dimensionality reduction mapping to the shape descriptor vectors and displaying the mapped shape descriptor vectors on a display.

16. A method according to claim 1 wherein said segmentation is performed automatically by a segmentation algorithm which in addition returns a value representing the confidence of the segmentation of each of said plurality of said repetitive signal features, and the method further comprises the steps of:

selecting a proportion of said repetitive signal features having low segmentation confidence values and displaying said repetitive signal features, while providing for their manual segmentation by reference to the display.

17. A method according to claim 1 wherein said segmentation is performed automatically by a segmentation algorithm, and the method further comprises the steps of:

selecting one or more of said repetitive signal features whose shape descriptor parameters depart from predefined normality by more than a predetermined threshold, and displaying said repetitive signal features, while providing for their manual segmentation by reference to the display.

18. A method according to claim 17 wherein said step of selecting one or more of said repetitive signal features whose shape descriptor parameters depart from predefined normality by more than a predetermined threshold comprises calculating the distance in shape descriptor vector space from the nearest shape descriptor vector of a predefined normal set, and comparing that to a predefined threshold.

19. A method according to claim 1, comprising the steps of:

selecting one or a group of waveforms from a first period in the biomedical signal and defining a first template based on the shape of said one or more waveforms, comparing subsequent waveforms in the signal to said template and generating a second template based on the shape of subsequent waveforms if their shape varies from the template by more than a predefined threshold, repeating the comparing and template generating steps until the end of the biomedical signal, and displaying said templates while providing for their manual segmentation by reference to the display.

20. A method according to claim 18 wherein for each of said selected one or more of said repetitive signal features whose shape descriptor parameters depart from predefined normality by more than a predetermined threshold, further of the signal features having similar shape descriptor parameters are selected to form respective groups of signal features, for each of said groups a template is defined based on the shape of the signal features in the group, and said template is displayed while providing for its manual segmentation by reference to the display.

21. A method according to claim 19 wherein the group comprises signal features whose shape descriptor vectors are within a predetermined distance of each other in shape descriptor vector space.

22. A method according to claim 21 wherein the template is defined by repetitive signal features in the biomedical signal whose shape descriptor vector is closest to the centre of the group in shape descriptor vector space.

23. A method according to claim 19, wherein the groups each comprise signal features which are temporally localised in the biomedical signal.

24. A method according to claim 22 wherein said template is defined by taking an average of plural adjacent repetitive signal features in the biomedical signal whose shape descriptor vector is closest to the centre of the group in shape descriptor vector space.

25. A method according to claim 24 wherein said plural adjacent repetitive signal features are subjected to dynamic time warping to match signal features.

26. A method according to claim 19 further comprising applying the manual segmentation of the template to the signal features of the corresponding group.

27. A method according to claim 26 further comprising the step of applying reverse dynamic time warping to apply the manual segmentation of the template to the signal features of the corresponding group.

28. A method according to claim 1 further comprising the step of retraining the segmentation algorithm using the manual segmentation of the template.

29. A method according to claim 20 wherein the group comprises signal features whose shape descriptor vectors are within a predetermined distance of each other in shape descriptor vector space.

30. A method according to claim 22 wherein the template is defined by repetitive signal features in the biomedical signal whose shape descriptor vector is closest to the centre of the group in shape descriptor vector space.

31. A method according to claim 20 wherein the groups each comprise signal features which are temporally localised in the biomedical signal.

32. A method according to claim 24 wherein said template is defined by taking an average of plural adjacent repetitive signal features in the biomedical signal whose shape descriptor vector is closest to the centre of the group in shape descriptor vector space.

33. A method according to claim 28 wherein said plural adjacent repetitive signal features are subjected to dynamic time warping to match signal features.

35. A method according to claim 32 further comprising the step of applying reverse dynamic time warping to apply the manual segmentation of the template to the signal features of the corresponding group.

34. A method according to 20 further comprising applying the manual segmentation of the template to the signal features of the corresponding group.

36. A method according to 20 further comprising the step of retraining the segmentation algorithm using the manual segmentation of the template.