Translating ai algorithms from 12-lead clinical ECGS to portable and consumer ECGS with fewer leads

1 . A system comprising:

at least one memory; and at least one processor coupled to the at least one memory, the system configured to cause the at least one processor to execute instructions stored in the at least one memory to:

receive first electrocardiogram (ECG) data associated with a plurality of subjects and an electrocardiogram configuration including a predetermined set of leads and a time interval, the predetermined set of leads defining a first number and configuration of leads, the first electrocardiogram data comprising, for each lead included in the predetermined set of leads, voltage data associated with at least a portion of the time interval,

identify a first artificial intelligence model for a particular cardiac disease state;

train the first artificial intelligence model on the first ECG data;

receive second ECG data derived from a particular device, the second ECG data being associated with a second number of leads, the second number of leads being fewer leads than the first number of leads;

select, from the predetermined set of leads, a training lead or combination of leads that most closely resembles the at least one lead of the particular device, at least in part by iteratively replacing data from one or more leads of the predetermined set of leads with ECG data associated with the second number of leads and evaluating performance of each iteration of the first artificial intelligence model;

fine-tune the first artificial intelligence model by iteratively identifying one or more frozen layers and one or more unfrozen layers, freezing the one or more frozen layers, and retraining the unfrozen layers, wherein subsequent iterations comprise unfreezing upstream layers from a last frozen layer;

train a second artificial intelligence model using a training dataset comprising ECG data including the predetermined set of leads, the second artificial intelligence model being configured to generate artificial ECG data corresponding to each lead of the predetermined set of leads based on the training lead or combination of leads;

input the second ECG data into the trained second artificial intelligence model;

translate, by the second artificial intelligence model, the second ECG data into artificial ECG data corresponding to each of the leads of the predetermined set of leads; and

process, by the first artificial intelligence model, the artificial ECG data to generate one or more predictions regarding the particular cardiac disease state.

2 . The system of claim 1 , wherein the second artificial intelligence model is trained using a generative adversarial network.

3 . The system of claim 1 , wherein the one or more predictions include a likelihood of experiencing a cardiac disease state within a predetermined period of time from when the particular device generates the second ECG data.

4 . The system of claim 1 , wherein the system further is configured to cause the at least one processor to execute instructions stored in the at least one memory to:

display a notification relating to the one or more predictions on a display screen of the particular device.

5 . The system of claim 4 , wherein the particular device includes a wearable device, and wherein the wearable device is wearable on a wrist or around a chest of a user.

6 . The system of claim 1 , wherein the second number of leads is one lead.

7 . The system of claim 1 , wherein the predetermined set of leads comprises 12 leads.

8 . A computer program product, the computer program product comprising instructions stored on a non-transitory computer readable medium to cause at least one processor on a computer to:

receive first electrocardiogram (ECG) data associated with a plurality of subjects and an electrocardiogram configuration including a predetermined set of leads and a time interval, the predetermined set of leads including a first number of leads, the first electrocardiogram data comprising, for each lead included in the predetermined set of leads, voltage data associated with at least a portion of the time interval,

identify a first artificial intelligence model for a particular cardiac disease state;

train the first artificial intelligence model on the first ECG data;

receive second ECG data derived from a particular device, the second ECG data being associated with a second number of leads, the second number of leads being fewer leads than the first number;

select, from the predetermined set of leads, a training lead that most closely resembles at least one lead of the particular device, at least in part by iteratively replacing data from one or more leads of the predetermined set of leads with ECG data associated with the second number of leads and evaluating performance of each iteration of the first artificial intelligence model;

fine-tune the first artificial intelligence model by iteratively identifying one or more frozen layers and one or more unfrozen layers, freezing the one or more frozen layers, and retraining the unfrozen layers, wherein subsequent iterations comprise unfreezing upstream layers from a last frozen layer;

train a second artificial intelligence model using a training dataset comprising ECG data including the predetermined set of leads, the second artificial intelligence model being configured to generate artificial ECG data corresponding to each lead of the predetermined set of leads based on the training lead;

record a pairing of the particular device and the second artificial intelligence model to a database of a compatibility service;

input the second ECG data into the trained second artificial intelligence model;

translate, by the second artificial intelligence model, the second ECG data into artificial ECG data corresponding to the leads of the predetermined set of leads; and

process, by the first artificial intelligence model, the artificial ECG data to generate one or more predictions regarding the particular cardiac disease state.

9 . The computer program product of claim 8 , wherein the second artificial intelligence model is trained using a generative adversarial network.

10 . The computer program product of claim 8 , wherein the one or more predictions include a likelihood of experiencing a cardiac disease state within a predetermined period of time from when the particular device generates the second ECG data.

11 . The computer program product of claim 8 , wherein the instructions further cause the at least one processor to:

display a notification relating to the one or more predictions on a display screen of the particular device.

12 . The computer program product of claim 11 , wherein the particular device is wearable by a user.

13 . The computer program product of claim 11 , wherein the particular device is wearable on a wrist of a user.

14 . The computer program product of claim 8 , wherein the particular device is wearable around a chest of a user.

15 . The computer program product of claim 8 , wherein the second number of leads is one lead.

16 . The computer program product of claim 8 , wherein the standardized plurality of leads comprises 12 leads.