US 12,543,977 B2 · Machine learning in an artificial pancreas

Abstract

Machine learning in an artificial pancreas is described. An artificial pancreas system may include a wearable glucose monitoring device, an insulin delivery system, and a computing device. Broadly speaking, the wearable glucose monitoring device provides glucose measurements of a person continuously. The artificial pancreas algorithm, which may be implemented at the computing device, determines doses of insulin to deliver to the person based on a variety of aspects for the purpose of maintaining the person's glucose within a target range, as indicated by those glucose measurements. The insulin delivery system then delivers those determined doses to the person. As the artificial pancreas algorithm determines insulin doses for the person over time and effectiveness of the insulin doses to maintain the person's glucose level in the target range is observed, an underlying model of the artificial pancreas algorithm may be updated to better determine insulin doses.

Claims31

Dependent claims

A method comprising:

receiving historical glucose measurements provided by a wearable glucose monitoring device worn by a user over a sequence of days;
receiving other user data of the user from one or more sources different from the wearable glucose monitoring device, the other user data comprising one or more of food consumption data, activity data, or insulin doses that are correlated to the glucose measurements based on timestamps of the glucose measurements and the other user data;
training, using one or more machine learning techniques, an artificial pancreas algorithm to determine insulin doses for the user to thereby provide a trained artificial pancreas algorithm, wherein the training is based on the historical glucose measurements and the other user data of the user that are correlated to the glucose measurements based on the timestamps of the glucose measurements and the other user data;
predicting an occurrence of an event impacting a glucose level of the user using the trained artificial pancreas algorithm, wherein predicting the occurrence of the event comprises predicting a type of event from among a plurality of types of events;
and determining an insulin dose based at least in part on the predicted event.

2 more elements

A system comprising:

an artificial pancreas controller to:
receive historical glucose measurements provided by a wearable glucose monitoring device worn by a user over a sequence of days;
receive other user data of the user from one or more sources different from the wearable glucose monitoring device, the other user data comprising one or more of food consumption data, activity data, or insulin doses that are correlated to the glucose measurements based on timestamps of the glucose measurements and the other user data;
and train, using one or more machine learning techniques, an artificial pancreas algorithm to determine insulin doses for the user to thereby provide a trained artificial pancreas algorithm, wherein the training is based on the historical glucose measurements and the other user data of the user that are correlated to the glucose measurements based on the timestamps of the glucose measurements and the other user data;
and predict an occurrence of an event impacting a glucose level of the user using the trained artificial pancreas algorithm, wherein predicting the occurrence of the event comprises predicting a type of event from among a plurality of types of events;
and determine an insulin dose based at least in part on the predicted event.

3 more elements

One or more computer-readable storage media having instructions stored thereon that are executable by one or more processors to perform operations comprising:

receiving historical glucose measurements provided by a wearable glucose monitoring device worn by a user over a sequence of days;
receiving other user data of the user from one or more sources different from the wearable glucose monitoring device, the other user data comprising one or more of food consumption data, activity data, or insulin doses that are correlated to the glucose measurements based on timestamps of the glucose measurements and the other user data;
training, using one or more machine learning techniques, an artificial pancreas algorithm to determine insulin doses for the user to thereby provide a trained artificial pancreas algorithm, wherein the training is based on the historical glucose measurements and the other user data of the user that are correlated to the glucose measurements based on the timestamps of the glucose measurements and the other user data;
and predicting an occurrence of an event impacting a glucose level of the user using the trained artificial pancreas algorithm, wherein predicting the occurrence of the event comprises predicting a type of event from among a plurality of types of events;
and determining an insulin dose based at least in part on the predicted event.

2 more elements

Dependencies are inferred from the claim text.

Machine learning in an artificial pancreas

Life of this patent

Machine learning in an artificial pancreas

Life of this patent

Abstract

Claims31