Method of detecting abnormalities in ECG signals
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.
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;
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.
19. A system according to claim 18 , wherein the processing unit is located within the ECG device or the server.