Systems and methods of identity analysis of electrocardiograms
A set of training electrocardiograms (ECGs) for each of a plurality of subjects is processed using a machine learning model to generate an output for each training ECG of each of the plurality of subjects. Training ECGs for each subject are labeled with an identity of the subject. A machine learning model is trained by comparing the output generated for each training ECG to a corresponding label of the training ECG to generate an identity model to identify ECGs of a first subject of the plurality of subjects. A first ECG is received from an ECG sensor and input to the identity model, which generates an output indicating whether the first ECG corresponds to the first subject. In response to the output indicating that the first ECG does not correspond to the first subject, a condition that the first subject has or may develop is determined based on the output.
1. A method comprising:
processing training data comprising a set of training electrocardiograms (ECGs) for each of a plurality of subjects using a machine learning model to generate an output for each training ECG of each of the plurality of subjects, wherein for each of the plurality of subjects, the set of training ECGs for the subject is labeled with an identity of the subject;
training the machine learning model by comparing the output generated for each training ECG to a corresponding label of the training ECG to generate an identity model to identify ECGs of a first subject of the plurality of subjects;
inputting a first ECG into the identity model, the identity model to generate an output indicating whether the first ECG corresponds to the first subject; and
in response to the output indicating that the first ECG does not correspond to the first subject, determining based on the output, a condition that the first subject has or may develop.
2. The method of claim 1 , wherein training ECGs for each of the plurality of subjects other than the first subject indicate a condition, and the machine learning model comprises an encoder that encodes the training ECGs of the first subject into a first region of a latent space and encodes the training ECGs for each of the plurality of subjects other than the first subject into a respective region of the latent space.
3. The method of claim 2 , wherein generating the output indicating whether the first ECG corresponds to the first subject comprises:
encoding the first ECG into a first position within the latent space; and
determining whether the first position is within the first region of the latent space.
4. The method of claim 3 , wherein determining based on the output, the condition that the first subject has or may develop comprises:
determining a vector represented by a difference in the first position relative to a position within the latent space of each of the training ECGs for each of the plurality of other subjects.
5. The method of claim 2 , wherein the output indicating whether the first ECG corresponds to the first subject comprises a probability that the first ECG corresponds to the first subject.
6. The method of claim 5 , wherein the probability that the first ECG corresponds to the first subject is based on a proximity of the first position to the first region of the latent space.
7. The method of claim 2 , wherein the machine learning model comprises a variational autoencoder.
8. A system comprising:
an electrocardiogram (ECG) sensor;
a memory; and
a processing device operatively coupled to the memory, the processing device to:
process training data comprising a set of training ECGs for each of a plurality of subjects using a machine learning model to generate an output for each training ECG of each of the plurality of subjects, wherein for each of the plurality of subjects, the set of training ECGs for the subject is labeled with an identity of the subject;
train the machine learning model by comparing the output generated for each training ECG to a corresponding label of the training ECG to generate an identity model to identify ECGs of a first subject of the plurality of subjects;
receive, from the ECG sensor, a first ECG;
process the first ECG using the identity model, the identity model to generate an output indicating whether the first ECG corresponds to the first subject; and
in response to the output indicating that the first ECG does not correspond to the first subject, determine based on the output, a condition that the first subject has or may develop.
9. The system of claim 8 , wherein training ECGs for each of the plurality of subjects other than the first subject indicate a condition, and the machine learning model comprises an encoder that encodes the training ECGs of the first subject into a first region of a latent space and encodes the training ECGs for each of the plurality of subjects other than the first subject into a respective region of the latent space.
10. The system of claim 9 , wherein to generate the output indicating whether the first ECG corresponds to the first subject, the processing device is to:
encode the first ECG into a first position within the latent space; and
determine whether the first position is within the first region of the latent space.
11. The system of claim 10 , wherein to determine based on the output, the condition that the first subject has or may develop, the processing device is to:
determine a vector represented by a difference in the first position relative to a position within the latent space of each of the training ECGs for each of the plurality of other subjects.
12. The system of claim 9 , wherein the output indicating whether the first ECG corresponds to the first subject comprises a probability that the first ECG corresponds to the first subject.
13. The system of claim 12 , wherein the probability that the first ECG corresponds to the first subject is based on a proximity of the first position to the first region of the latent space.
14. The system of claim 9 , wherein the machine learning model comprises a variational autoencoder.
15. A non-transitory computer-readable medium having instructions stored thereon which, when executed by a processing device, cause the processing device to:
process training data comprising a set of training electrocardiograms (ECGs) for each of a plurality of subjects using a machine learning model to generate an output for each training ECG of each of the plurality of subjects, wherein for each of the plurality of subjects, the set of training ECGs for the subject is labeled with an identity of the subject;
train the machine learning model by comparing the output generated for each training ECG to a corresponding label of the training ECG to generate an identity model to identify ECGs of a first subject of the plurality of subjects;
input a first ECG into the identity model, the identity model to generate an output indicating whether the first ECG corresponds to the first subject; and
in response to the output indicating that the first ECG does not correspond to the first subject, determine based on the output, a condition that the first subject has or may develop.
16. The non-transitory computer-readable medium of claim 15 , wherein training ECGs for each of the plurality of subjects other than the first subject indicate a condition, and the machine learning model comprises an encoder that encodes the training ECGs of the first subject into a first region of a latent space and encodes the training ECGs for each of the plurality of subjects other than the first subject into a respective region of the latent space.
17. The non-transitory computer-readable medium of claim 16 , wherein to generate the output indicating whether the first ECG corresponds to the first subject, the processing device is to:
encode the first ECG into a first position within the latent space; and
determine whether the first position is within the first region of the latent space.
18. The non-transitory computer-readable medium of claim 17 , wherein to determine based on the output, the condition that the first subject has or may develop, the processing device is to:
determine a vector represented by a difference in the first position relative to a position within the latent space of each of the training ECGs for each of the plurality of other subjects.
19. The non-transitory computer-readable medium of claim 16 , wherein the output indicating whether the first ECG corresponds to the first subject comprises a probability that the first ECG corresponds to the first subject.
20. The non-transitory computer-readable medium of claim 19 , wherein the probability that the first ECG corresponds to the first subject is based on a proximity of the first position to the first region of the latent space.