SYNTHETIC DATA AUGMENTATION FOR ECG USING DEEP LEARNING
A method includes generating first electrocardiogram (ECG) data by adding synthetic noise to naturally occurring ECG data using a first deep neural network (DNN). The method further includes providing one of: (i) the first ECG data, or (ii) second ECG data including naturally occurring noise, to a second DNN. An output is generated by the second DNN indicating whether the second DNN received the first ECG data or the second ECG data.
1 . A computer-implemented method, comprising:
generating first electrocardiogram (ECG) data by adding synthetic noise to naturally occurring ECG data using a first deep neural network (DNN);
providing one of: (i) the first ECG data, or (ii) second ECG data including naturally occurring noise, to a second DNN; and
generating an output by the second DNN indicating whether the second DNN received the first ECG data or the second ECG data,
wherein the output is used to further train the first DNN, and
wherein the first DNN is configured to be used to generate training data to train a machine learning (ML) model to identify a cardiac irregularity in patient ECG data.
2 . The computer-implemented method of claim 1 , wherein the first DNN and the second DNN are co-trained, such that the first ECG data is used to train the second DNN and the output from the second DNN is used to train the first DNN.
3 . The computer-implemented method of claim 1 , wherein one of the first ECG data or the second ECG data is randomly provided to the second DNN.
4 . The computer-implemented method of claim 1 , wherein the first DNN uses ECG data without naturally occurring noise and randomly generated noise to generate the first ECG data.
5 . The computer-implemented method of claim 1 , wherein the output further comprises a confidence indicator.
6 . The computer-implemented method of claim 5 , wherein the first DNN uses both the confidence indicator and the indication of whether the second DNN received the first ECG data or the second ECG data to generate the first ECG data.
7 . The computer-implemented method of claim 1 , further comprising:
generating third ECG data comprising synthetic noise using the trained first DNN; and
training the ML model using the third ECG data.
8 . The computer-implemented method of claim 7 , further comprising:
identifying a cardiac irregularity in a patient using fourth ECG data and the trained ML model, wherein the fourth ECG data comprises naturally occurring ECG data from the patient and naturally occurring noise.
9 . The computer-implemented method of claim 8 , further comprising:
treating the patient based on the cardiac irregularity.
10 . A non-transitory computer-readable medium containing computer program code that, when executed by a computer processor, performs an operation comprising:
generating first electrocardiogram (ECG) data by adding synthetic noise to naturally occurring ECG data using a first deep neural network (DNN);
providing one of: (i) the first ECG data, or (ii) second ECG data including naturally occurring noise, to a second DNN; and
generating an output by the second DNN indicating whether the second DNN received the first ECG data or the second ECG data,
wherein the output is used to further train the first DNN, and
wherein the first DNN is configured to be used to generate training data to train a machine learning (ML) model to identify a cardiac irregularity in patient ECG data.
11 . The non-transitory computer-readable medium of claim 10 , wherein the first DNN and the second DNN are co-trained, such that the first ECG data is used to train the second DNN and the output from the second DNN is used to train the first DNN.
12 . The non-transitory computer-readable medium of claim 10 , wherein one of the first ECG data or the second ECG data is randomly provided to the second DNN.
13 . The non-transitory computer-readable medium of claim 10 , wherein the first DNN uses ECG data without naturally occurring noise and randomly generated noise to generate the first ECG data.
14 . The non-transitory computer-readable medium of claim 10 , wherein the output further comprises a confidence indicator, and wherein the first DNN uses both the confidence indicator and the indication of whether the second DNN received the first ECG data or the second ECG data to generate the first ECG data.
15 . The non-transitory computer-readable medium of claim 10 , the operation further comprising:
generating third ECG data comprising synthetic noise using the trained first DNN;
training the ML model using the third ECG data;
identifying a cardiac irregularity in a patient using fourth ECG data and the trained ML model, wherein the fourth ECG data comprises naturally occurring ECG data from the patient and naturally occurring noise; and
treating the patient based on the cardiac irregularity.
16 . A system, comprising:
a computer processor; and
a memory having instructions stored thereon which, when executed on the computer processor, performs an operation comprising:
generating first electrocardiogram (ECG) data by adding synthetic noise to naturally occurring ECG data using a first deep neural network (DNN);
providing one of: (i) the first ECG data, or (ii) second ECG data including naturally occurring noise, to a second DNN; and
generating an output by the second DNN indicating whether the second DNN received the first ECG data or the second ECG data,
wherein the output is used to further train the first DNN, and
wherein the first DNN is configured to be used to generate training data to train a machine learning (ML) model to identify a cardiac irregularity in patient ECG data.
17 . The system of claim 16 , wherein the first DNN and the second DNN are co-trained, such that the first ECG data is used to train the second DNN and the output from the second DNN is used to train the first DNN.
18 . The system of claim 16 , wherein the first DNN uses ECG data without naturally occurring noise and randomly generated noise to generate the first ECG data.
19 . The system of claim 16 , wherein the output further comprises a confidence indicator, and wherein the first DNN uses both the confidence indicator and the indication of whether the second DNN received the first ECG data or the second ECG data to generate the first ECG data.
20 . The system of claim 16 , the operation further comprising:
generating third ECG data comprising synthetic noise using the trained first DNN;
training the ML model using the third ECG data;
identifying a cardiac irregularity in a patient using fourth ECG data and the trained ML model, wherein the fourth ECG data comprises naturally occurring ECG data from the patient and naturally occurring noise; and
treating the patient based on the cardiac irregularity.