Systems and methods of analyte measurement analysis

1. A system for non-invasively predicting a level of an analyte comprising:

an electrocardiogram sensor; and

a processing device operatively coupled to the electrocardiogram sensor, wherein the processing device is to:

analyze electrocardiogram training data comprising an electrocardiogram of each of a plurality of subjects and one or more measured analyte levels associated with each electrocardiogram to determine an associated label for each electrocardiogram of the electrocardiogram training data, the associated label for each electrocardiogram of the electrocardiogram training data based on a regression line fitted to the one or more measured analyte levels associated with each electrocardiogram, wherein the associated label is determined from the regression line at a time of measurement of the associated electrocardiogram;

train a machine learning model by:

analyzing each electrocardiogram of the electrocardiogram training data to generate an output; and

comparing the output generated for each electrocardiogram of the electrocardiogram training data to the associated label for each electrocardiogram of the electrocardiogram training data to update the machine learning model using backpropagation, wherein one or more weight matrices of the machine learning model are adjusted based on a confidence interval associated with the associated label for each electrocardiogram of the electrocardiogram training data;

receive electrocardiogram data of a subject from the electrocardiogram sensor; and

apply the machine learning model to the received electrocardiogram data to determine an indication of a measured analyte level of the subject based on the electrocardiogram data.

2. The system of claim 1 , wherein the received electrocardiogram data comprises an electrocardiogram signal measured over multiple heartbeats of the subject.

3. The system of claim 2 , wherein applying the machine learning model comprises pre-processing the electrocardiogram signal to generate an average heartbeat over the multiple heartbeats of the subject.

4. The system of claim 1 , wherein the electrocardiogram sensor comprises a 2 lead or 3 lead electrocardiogram sensor.

5. The system of claim 1 , wherein the machine learning model is one of a convolutional neural network, a recurrent neural network, or a combination of a convolutional and a recurrent neural network.

6. The system of claim 1 , wherein the analyte is one of potassium, magnesium, or calcium.

7. The system of claim 1 , wherein the indication of the measured level of the analyte indicates one of an estimate of concentrations of the measured analyte in the subject or an indication of a classification of the measured analyte level as high or normal.

8. The system of claim 1 , wherein the processing device is part of a mobile device comprising a display screen, and wherein the processing device is further to cause the display screen to display the indication of the level of the analyte on the display screen.

9. A method, comprising:

receiving, by a processing device, electrocardiogram training data comprising a plurality of electrocardiograms and one or more analyte measurements associated with each of the electrocardiograms;

determining, by the processing device, training labels for the plurality of electrocardiograms of the electrocardiogram training data based at least in part on a regression line fitted to the one or more analyte measurements associated with each of the electrocardiograms, wherein the training labels are determined from the regression line at a time of measurement of the associated electrocardiogram of the plurality of electrocardiograms;

training a machine learning model by:

analyzing each electrocardiogram of the electrocardiogram training data to generate an output; and

comparing, by the processing device, the output generated for each electrocardiogram of the electrocardiogram training data to the associated label for each electrocardiogram of the electrocardiogram training data to update the machine learning model using back propagation, wherein one or more weight matrices of the machine learning model are adjusted based on a confidence interval associated with the associated label for each electrocardiogram of the electrocardiogram training data.

10. The method of claim 9 , further comprising:

receiving new electrocardiogram data of a subject from an electrocardiogram sensor, the new electrocardiogram data not in the electrocardiogram training data;

inputting the new electrocardiogram data into the machine learning model trained on the electrocardiogram training data to generate a new output; and

determining an indication of an analyte concentration in the subject based on the new output of the machine learning model.

11. The method of claim 10 , further comprising determining a classification of the analyte concentration in the subject as high or low.

12. The method of claim 9 , wherein determining a particular training label of the training labels for the plurality of electrocardiograms comprises:

determining, for a particular electrocardiogram in the electrocardiogram training data, an estimated analyte concentration at a time the particular electrocardiogram data was generated based on the received electrocardiogram training data and the one or more analyte measurements associated with the particular electrocardiogram; and

determining whether the particular electrocardiogram of the electrocardiogram training data satisfies a threshold to be labeled as a first classification.

13. The method of claim 9 , wherein the regression line is based on a Gaussian Process Regression analysis generated based on the one or more analyte measurements associated with each of the plurality of electrocardiograms.

14. The method of claim 12 , wherein determining the particular training label of the training labels for the plurality of electrocardiogram further comprises:

determining that a confidence interval around the estimated analyte concentration of the particular electrocardiogram in the electrocardiogram training data is larger than a threshold; and

removing the particular electrocardiogram from the plurality of electrocardiograms in response to the estimated analyte concentration being higher than a threshold.

15. The method of claim 9 , wherein the analyte is one of potassium, magnesium, or calcium.