Automatic method to delineate or categorize an electrocardiogram
A method for computerizing delineation and/or multi-label classification of an ECG signal, includes: applying a neural network to the ECG whereby labelling the ECG, and optionally displaying the labels according to time, optionally with the ECG signal.
1. A computerized-method for classification of an electrocardiogram (ECG) signal obtained from a patient using a neural network, the computerized-method comprising:
training the neural network with a dataset of pre-characterized ECG signals to generate a trained neural network;
receiving the ECG signal sampled at a plurality of time points;
analyzing the ECG signal at the plurality of time points using the trained neural network to compute anomaly scores based on the ECG signal to detect anomalies;
assigning an anomaly label to the ECG signal for a detected anomaly based on the anomaly scores;
defining an interval of time for the detected anomaly of the ECG signal according to time and based on the anomaly scores, the interval of time comprising length and location information in the ECG signal for the detected anomaly; and
causing display of a representation of the ECG signal, the anomaly label, and the interval of time.
2. The computerized-method of claim 1 , further comprising denoising and removing a baseline of the ECG signal.
3. The computerized-method of claim 1 , further comprising determining a location of additional anomalies in the ECG signal.
4. The computerized-method of claim 1 , further comprising assigning one or more additional anomaly labels to the ECG signal for at least one time point of the plurality of time points.
5. The computerized-method of claim 1 , further comprising expressing the ECG signal at a selected frequency.
6. The computerized-method of claim 1 , further comprising causing display of the anomaly scores.
7. The computerized-method of claim 1 , further comprising assigning one or more additional anomaly labels to the ECG signal based on the anomaly scores for one or more additional detected anomalies,
wherein causing display comprises causing display of the one or more additional anomaly labels.
8. The computerized-method of claim 1 , wherein assigning the anomaly label to the ECG signal comprises assigning the anomaly label of atrial fibrillation to the ECG signal.
9. The computerized-method of claim 1 , wherein:
the neural network expresses the ECG signal as a matrix of size m×n with m being a number of leads fixed and at a predefined frequency and n being a number of samples.
10. The computerized-method of claim 1 , further comprising expressing a plurality of detected anomalies as a vector of size q, with q being a number of anomalies to identify.
11. The computerized-method of claim 7 , wherein at least one of the one or more additional anomaly labels correspond to a normal label.
12. The computerized-method of claim 7 , further comprising defining one or more additional intervals of time for the one or more additional detected anomalies according to time and based on the anomaly scores,
wherein causing display comprising causing display of the one or more additional intervals of time.
13. A system for classification of an electrocardiogram (ECG) signal obtained from a patient using a neural network, the system comprising at least one processor to:
train the neural network with a dataset of pre-characterized ECG signals to generate a trained neural network;
receive the ECG signal sampled at a plurality of time points;
analyze the ECG signal at the plurality of time points using the trained neural network to compute anomaly scores based on the ECG signal to detect anomalies;
assign an anomaly label to the ECG signal for a detected anomaly based on the anomaly scores;
define an interval of time for the detected anomaly of the ECG signal according to time and based on the anomaly scores, the interval of time comprising length and location information in the ECG signal for the detected anomaly; and
cause display of a representation of the ECG signal, the anomaly label, and the interval of time.
14. The system of claim 13 , wherein the at least one processor is configured to denoise and remove a baseline of the ECG signal.
15. The system of claim 13 , wherein the at least one processor is configured to determine a location of additional anomalies in the ECG signal.
16. The system of claim 13 , wherein the at least one processor is further configured to assign one or more additional anomaly labels to the ECG signal for at least one time point of the plurality of time points.
17. The system of claim 13 , wherein the at least one processor is configured to express the ECG signal at a selected frequency.
18. The system of claim 13 , wherein the at least one processor is configured to cause display of the anomaly scores.
19. The system of claim 13 , wherein the at least one processor is further configured to assign one or more additional anomaly labels to the ECG signal based on anomaly scores for one or more additional detected anomalies, and to cause display of the one or more additional anomaly labels with the representation of the ECG signal.
20. The system of claim 13 , wherein the anomaly label comprises the anomaly label of atrial fibrillation.
21. The system of claim 13 , wherein the at least one processor is configured to express the ECG signal as a matrix of size m×n with m fixed and at a predefined frequency.
22. The system of claim 13 , wherein the at least one processor is configured to express one or more detected anomalies as a vector of size q, with q being a number of anomalies to identify.
23. The system of claim 19 , wherein at least one of the additional anomaly labels correspond to a normal label.
24. A programmed routine comprising instructions that, when executed by at least one processor, cause the at least one processor to:
train the neural network with a dataset of pre-characterized ECG signals to generate a trained neural network;
receive the ECG signal sampled at a plurality of time points;
analyze the ECG signal at the plurality of time points using the trained neural network to compute an anomaly scores based on the ECG signal to detect anomalies;
assign an anomaly label to the ECG signal for a detected anomaly based on the anomaly scores;
define an interval of time for the detected anomaly of the ECG signal according to time and based on the anomaly scores, the interval of time comprising length and location information in the ECG signal for the detected anomaly; and
cause display of a representation of the ECG signal, the anomaly label, and the interval of time.
25. The programmed routine of claim 24 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to denoise and remove a baseline of the ECG signal.
26. The programmed routine of claim 24 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to determine a location of additional anomalies in the ECG signal.
27. The programmed routine of claim 24 , further comprising instructions that, when executed by the at least one processor, cause the at least one processor to assign one or more additional anomaly labels to the ECG signal for at least one time point of the plurality of time points.
28. The programmed routine of claim 27 , wherein at least one of the additional anomaly labels correspond to a normal label.