Automatic method to delineate or categorize an electrocardiogram
Disclosed is a method for computerizing delineation and/or multi-label classification of an ECG signal, including: applying a neural network to the ECG, labelling the ECG, and optionally displaying the labels according to time with the ECG signal.
1. A computerized method for detecting abnormalities based on a cardiac signal obtained from a patient, the computerized-method comprising:
obtaining the cariac signal corresponding to a plurality of heart beats of the patient and comprising a plurality of time points;
generating a first matrix representative of the caridac signal;
applying the first matrix to a first neural network to generate a delineation matrix, the delineation matrix comprising for each on e of the plurality of time points a plurality of delineation scores, each delineation score corresponding to one of a plurality of wave types at one of the plurality of time points;
determining whether one or more wave types of the plurality of wave types are present at each time point of the plurality of time points by comparing the plurality of delineation scores to a delineation threshold value;
calculating at least one measurement corresponding to the cardiac signal from the delineation matrix based on the determination of whether one or more wave types are present at each time point;
applying the cardiac signal and the at least one measurement to a second neural network to generate at least one classification vector, the at least one classificaiton vector comprising a plurality of classification scores corresponding to a presence of one or more abnormality types;
determining whether one or more abnormality types arer presnet by comparing each one of the plurality of classification scores to a classification threshold value; and
generating information to indicate the presence of the one or more abnormality types based on determining that the one or more abnormality types are present.
2. The computerized method of claim 1 , wherein the first matrix is a matrix of size m×n where “m” is a number of leads of a cardiac sensor used to obtain the cardiac signal and “n” is a number of time points of the plurality of time points.
3. The computerized method of claim 1 , wherein the plurality of wave types comprises P-waves, QRS complexed, and T-waves.
4. The computerized method of claim 1 , wherein the delineation matrix is a matrix of size p×n where “p” is a number of wave types of the plurality of wave types and “n” is a number of time points of the plurality of time points.
5. The computerized method of claim 1 , wherein calculating at least one measurement corresponding to the cardiac signal comprises determining a beginning time point of the plurality of time points and an ending time point of the plurality of time points for at least one wave of the plurality of waves determined to be present.
6. The computerized method of claim 1 , wherein calculating at least one measurement corresponding to the cardiac signal comprises determining one or more of a P duration, PR interval, QRS duration, or QT interval.
7. The computerized method of claim 1 , wherein the at least one classification vector is a vector of size q where “q” is a number of abnormality types corresponding to the plurality of classification scores.
8. The computerized method of claim 1 , wherein each one of the plurality of delineation scores and each one of the plurality of classification scores are between the numbers 0 and 1.
9. The computerized method of claim 1 , further comprising denoising the cardiac signal.
10. The computerized method of claim 9 , further comprising removing the baseline frequency of the cardiac signal and expressing the cardiac signal at a chosen frequency.
11. The computerized method of claim 1 , wherein at least one of the first neural network or the second neural network is trained using at least one training cardiac signal with known parametsers by modifying the at least one of the first neural network or the second neural network to decrease a cost function based on the known parameters of the at least one training cardiac signal.
12. A system for detyecting abnormalities based on a cardiac signal obtained from a patient, the system comprising:
at least one server configured to:
obtain the cardiac signal corresponding to a plurality of heart beats of the patient and comprising a plurality of time points:
generate a first matrix representative of the cardiac signal;
apply the first matrix to a first neural network to generate a delineation matrix, the delineation matrix comprising for each one of the plurality of time points a plurality of delineation scores, each delineation score corresponding to one of a plurality of wave types at one of the plurality of time points;
determine whether one or more wave types of the plurality of wave types are present at each time point of the plurality of time points by comparing the plurality of delineation scores to a delineation threshold value;
calculate at least one measurement corresponding to the cardiac signal from the delineation matrix based on the determination of whether one or more wave types are present at each time point;
apply the cardiac signal and the at least one measurement to a second neural network to generate at least one classification vector, the at least one classification vector comprising a plurality of classification scores correspoinding to a presence of one or more abnormality types;
determine whether one or more abnormality types are present by comparing each one of the plurality of classification scores to a classification threshold value; and
generate information to indicate the presence of the one or more abnormality types based on determining that the one or more abnormality types are present.
13. The system of claim 12 , wherein the first matrix is a matrix of size m×n where “m” is anumber of leads of a cardiac sensor used to obtain the cardiac signal and “n” is a number of time points of the plurality of time points.
14. The system of claim 12 , wherein the plurality of wave types comprises P-waves, QRS complexe, and T-waves.
15. The system of claim 12 , wherein the delineation matrix is a matrix of size p×n where “p” is a number of wave types of the plurality of wave types and “n” is a number of time points of the plurality of time points.
16. The system of claim 12 , wherein the at least one server is further configured to determine a beginning time point of the plurality of time points and an ending time point of the plurality of time points for at least one wav e of the plurality of waves determined to be present.
17. The system of claim 12 , wehrein the at least one server is further configured to determine at least one of a P duration, PR interval, QRS duration, or QT interval.
18. The system of claim 12 , wherein the at least one classification vector is a vector of size q where “q” is a number of abnormality types corresponding to the plurality of classification scores.
19. The system of claim 12 , wherein each one of the plurality of delineation scores and each ione of hte plurality of classification scores are between the number 0 and 1.
20. The system of claim 12 , wherein the at least one server is further configured to denoise the cardial signal.
21. The system of claim 20 , wherein the at least one server is further configured to remove the baseline frequency of the cardiac signal and express the cardiac signal at a chosen frequency.
22. The system of claim 12 , wherein the at least one of the first neural network or the second neural network is trained using at least one training cardiac signal with known parameters by modifying the at least one of the first neural network or the secvond neural network to decrease a cost function based on the know parameters of the at least one training cardiac signal.
23. A programmed routine for use with a computer for detecting abnormalities based on a cardiac signal obtained from a patient, the programmed routine comprising instructions that when executed:
obtain the cardiac signal corresponding to a plurality of heart beats of the patient and comprising a plurality of time points;
generate a first matrix representative of the cardiac signal;
apply the first matrix to a first neural network to generate a delineation matrix, the delineation matrix comprising for each one of the plurality of time points a plurality of delineation scores, each delineation score corresponding to one of a plurality of wave types at one of the plurality of time points;
determine whether one or more wave types of the plurality of wave types are present at each time point of the plurality of time points by comparing the plurality of delineation scores to a delineation threshold value;
calculate at least one measurement corresponding to the cardiac signal from the delineation matrix based on the determination of whether one or more wave types are present at each time point;
apply the cardiac signal and the at least one measurement to a second neural network to generate at least one classification vector, the at least one classification vector comprisng a plurality of classification scores corresponding to a presence of one or more abnormality types;
determine whether one or more abnormality types are present by comparing each one of the plurality of classification scores to a classification threshold value; and
generate information to indicate the presence of the one or more abnormality types based on determining that the one or more abnormality types are present.
24. The programmed routine of claim 23 , wherein at least one of the first neural network or the second neural network is trained using at least one training cardiac signal with known parameters by modifying the at least one of the first neural network or the second neural network to decrease a cost function based on the known parameters of the at least one training cardiac signal.