| CPC G16H 50/30 (2018.01) [G16H 10/60 (2018.01); G16H 20/10 (2018.01); G16H 20/60 (2018.01); G16H 50/20 (2018.01)] | 15 Claims |
|
1. A computer-implemented method for training a machine learning model for predicting metabolic values, the method comprising:
sensing an individual's glucose levels by a continuous glucose monitoring (CGM) device over a period of time;
receiving the individual's glucose levels collected by the CGM device over the period of time;
determining a first glycemia risk index (GRI) value based on a first amount of time the individual is hypoglycemic during the period of time and a second amount of time the individual is hyperglycemic during the period of time;
determining a time in range (TIR) value of the individual's glucose level, wherein the determined TIR value is based on an amount of time the individual's glucose level is within a threshold band over the time period, wherein the threshold band is generated based on the lifestyle, habits, and medical test results of the individual;
receiving historical metabolic values for an individual having a first medical condition, wherein the individuals historical metabolic values are associated with heart rates, heart related values, ketone values, weight values, cortisol values, hormone levels, body electrical values, repertory values, and the glucose levels collected by the CGM device;
providing a first subset of the historical metabolic values, supplementary variables, and supplementary conditions to a machine learning model to train a generative machine learning model, wherein the first subset of the historical metabolic values includes the CGM values over the period of time, wherein the supplementary variables include the TIR value and GRI value, and wherein the supplementary condition is an anticipated dosage, anticipated time for taking a given medication, anticipated example exercise, or an anticipated example food to be consumed by the given individual;
generating a first predicted metabolic value based on the first subset of the historical metabolic values, wherein generating the first predicted metabolic value based on the first subset of the historical metabolic values includes generating the first predicted metabolic value at a first interval;
calculating a root mean square error (RMSE) between the first predicted metabolic value and a corresponding actual metabolic value of a second subset of the historical metabolic values;
training the generative machine learning model to minimize the RMSE, wherein the generative machine learning model is trained to learn the distribution of the historical metabolic values, the generative machine learning model is a statistical model of the joint probability distribution on the historical metabolic values and the predicted metabolic values, wherein the generative machine learning model is used to generate random instances of the historical metabolic values and may select a given historical metabolic value from the random instances that corresponds to a highest probability of occurring, wherein the generative machine learning model determines a conditional probability of a target metabolic value based on historical metabolic values, and wherein the generative machine learning model learns patterns and structure of training data that includes the historical metabolic values to generate new data that has similar characteristic;
generating a trained generative machine learning model based on the training, wherein the trained generative machine learning model outputs recommendations for causing the predicted metabolic value to comply with a metabolic value goal or range;
receiving a trained generative machine learning model output;
calculating a timing and a dosing amount of insulin for the individual based on the trained generative machine learning model output; and
automatically administering, via an insulin pump, based on the trained generative machine learning model output, at the timing, the dosing amount of insulin to the individual.
|