METHOD OF DETECTING ABNORMALITIES IN ECG SIGNALS

• US 20190059763A1
• Filed: 08/25/2017
• Published: 02/28/2019
• Est. Priority Date: 08/25/2017
• Status: Active Grant

• Current Assignee: Cambridge Heartwear Limited
• Original Assignee: Cambridge Heartwear Limited
• Inventors: Shakur, Rameen, Tan, Levin Chun Kiat, Cipolla, Roberto
• Granted Patent: US 10,602,942 B2
• Time in Patent Office: Days
• CPC Class Codes: A61B 5/0022 Monitoring a patient using a global network, e.g. telephone networks, internet
• CPC Class Codes: A61B 5/02438 with portable devices, e.g. worn by the patient
• CPC Class Codes: A61B 5/0245 by using sensing means generating electric signals, i.e. ECG signals
• CPC Class Codes: A61B 5/316 Modalities, i.e. specific diagnostic methods
• CPC Class Codes: A61B 5/318 Heart-related electrical modalities, e.g. electrocardiography [ECG]
• CPC Class Codes: A61B 5/349 Detecting specific parameters of the electrocardiograph cycle
• CPC Class Codes: A61B 5/352 Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
• CPC Class Codes: A61B 5/361 Detecting fibrillation
• CPC Class Codes: A61B 5/72 Signal processing specially adapted for physiological signals or for diagnostic purposes character recognition, recognising digital ink, document-oriented image-based pattern recognition G06V30/00; recognition or understanding of postal images G06V30/424
• CPC Class Codes: A61B 5/725 using specific filters therefor, e.g. Kalman or adaptive filters A61B5/316 take precedence
• 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: A61B 5/7485 Automatic selection of region of interest
• CPC Class Codes: G16H 40/67 for remote operation
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

We disclose herein a method of detecting abnormalities in electrocardiogram (ECG) signals, the method comprising receiving a set of ECG signals from an ECG device; amplifying only the peaks of at least some of the set of ECG signals to produce ECG beat markings from which a heart rate is derivable to detect an irregular rhythm between at least two ECG beats; extracting a single ECG beat from the set of ECG signals from the ECG device by using said ECG beat markings; feeding the extracted single ECG beat into a first neural network; producing, at the first neural network, a compact representation of the extracted single ECG signal so as to generate a feature extraction output; and using, at a second neural network, the feature extraction output from the first neural network to generate a score associated with the abnormalities in the ECG signals.

Claims

1. A method of detecting abnormalities in electrocardiogram (ECG) signals, the method comprising:

receiving a set of ECG signals from an ECG device;

amplifying only the peaks of at least some of the set of ECG signals to produce ECG beat markings from which a heart rate is derivable to detect an irregular rhythm between at least two ECG beats;

extracting a single ECG beat from the set of ECG signals from the ECG device by using said ECG beat markings;

feeding the extracted single ECG beat into a first neural network;

producing, at the first neural network, a compact representation of the extracted single ECG signal so as to generate a feature extraction output; and

using, at a second neural network, the feature extraction output from the first neural network to generate a score associated with the abnormalities in the ECG signals.

2. A method according to claim 1, wherein said irregular rhythm is detected from an irregular R-R interval flag, and wherein the score generated at the second neural network determines the presence or absence of a P-wave in the ECG signals.

3. A method according to claim 2, wherein the method further comprises combining the irregular R-R interval flag with the score from the second neural network for identifying abnormalities in the ECG signals.

4. A method according to claim 1, wherein said amplifying only the peaks of at least some of the set of ECG signals is performed by a matched filtering technique.

5. A method according to claim 4, wherein the matched filtering technique reduces a base line wander of the set of ECG signals.

6. A method according to claim 4, wherein the matched filtering technique uses an algorithm comprising a first derivative of a Gaussian and/or a second derivative of a Gaussian.

7. A method according to claim 4, comprising applying an additive thresholding technique on the matched filtered signal which generates said ECG beat markings.

8. A method according to claim 1, wherein said extracting the single ECG beat is conducted by cutting between two consecutive R-R intervals.

9. A method according to claim 8, further comprising scaling the extracted single beat through resampling so that it fits into a predetermined number of samples.

10. A method according to claim 1, wherein at least one of the first and second neural networks is a feed forward artificial neural network.

11. A method according to claim 1, wherein the first neural network is trained as an auto-encoding neural network having a pinch point, and wherein said compact representation of the single ECG beat is obtained by splitting the auto-encoding neural network at the pinch point.

12. A method according to claim 11, wherein said compact representation of the single ECG beat is obtained by using a front end of the auto-encoding neural network.

13. A method according to claim 11, wherein an output from the pitch point of the auto-encoding network is a feature vector of a predetermined size, and optionally wherein the feature vector generates a plurality of clusters, wherein features associated with abnormalities are grouped in at least some clusters.

14. A method according to claim 1, wherein the second neural network is a trained classifier network which generates the score associated with the abnormalities relating to the presence or absence of P-wave in ECG signals.

15. A method according to claim 14, further comprising training a support vector machine with a Gaussian algorithm to compare the results from the classifier network.

16. A method according to claim 1, wherein the second neural network is a classifier network which determines any one of:

(5) atrial fibrillation (AF);

(6) inverted T-waves;

(7) deep Q-waves; and

/or

(8) deep S-waves.

17. A method according to claim 1, further comprising a plurality of second neural networks, each determining a separate symptom from the following list:

(5) atrial fibrillation (AF);

(6) inverted T-waves;

(7) deep Q-waves; and

/or

(8) deep S-waves.

18. A system for detecting abnormalities in electrocardiogram (ECG) signals, the system comprising:

an ECG device configured to obtain a set of ECG signals;

a mobile device configured to receive ECG signals from the ECG device;

a server configured to receive ECG data from the mobile device; and

a processing unit configured to process the set of ECG signals from the ECG device and detect abnormalities in the ECG signals.

19. A system according to claim 18, wherein the processing unit is located within the ECG device or the server.

20. A system according to 18, wherein the processing unit comprises:

an irregular rhythm detector for amplifying only the peaks of at least some of the set of ECG signals to produce ECG beat markings from which a heart rate is derivable to detect an irregular rhythm between at least two ECG beats;

a beat extractor for extracting a single ECG beat from the set of ECG signals from the ECG device by using said ECG beat markings;

a first neural network for receiving the extracted single ECG beat and for producing a compact representation of the extracted single ECG signal so as to generate a feature extraction output; and

a second neural network for using the feature extraction output from the first neural network to generate a score associated with the abnormalities in the ECG signals.