Micro models and layered prediction models for estimating sensor glucose values and reducing sensor glucose signal blanking

Patent metadata

Field	Value
Publication number	US-12161464-B2
Application number	US-202117163149-A
Country	US
Kind code	B2
Filing date	Jan 29, 2021
Priority date	Jan 22, 2021
Publication date	Dec 10, 2024
Grant date	Dec 10, 2024

Abstract

Methods, systems, and devices for improving continuous glucose monitoring (“CGM”) are described herein. More particularly, the methods, systems, and devices describe applying micro machine learning models to generate predicted sensor glucose values. The system may use the predicted sensor glucose values to display a sensor glucose value to a user. The layered models may generate more reliable sensor glucose predictions across many scenarios, leading to a reduction of sensor glucose signal blanking. The methods, systems, and devices described herein further comprise applying a plurality of micro model to estimate sensor glucose values under outlier conditions. The system may prioritize the models that are trained for certain outlier conditions when the system detects those outlier condition based on the sensor data.

First claim

What is claimed is:

1. A sensor device for applying micro machine learning models to reduce sensor glucose signal blanking, the sensor device comprising:
memory configured to store a plurality of machine learning models, wherein each machine learning model of the plurality of machine learning models is trained to predict a sensor glucose value under a particular outlier condition using training data comprising clinical data on the particular outlier condition; and
a processor configured to:
receive continuous glucose monitoring (CGM) sensor data;
identify a signature of input features in the CGM sensor data;
adjust weights of the plurality of machine learning models based on the signature of input features in the CGM sensor data;
receive an output from the plurality of machine learning models, the output indicating a predicted sensor glucose value based on weighting outputs of the plurality of machine learning models using the adjusted weights; and
cause displaying, on a display interface, the output indicating the predicted sensor glucose value such that the sensor glucose signal blanking is reduced.

2. The sensor device of claim 1 , wherein identifying the signature of input features in the CGM sensor data comprises matching a combination of input features in the CGM sensor data with a predetermined signature of input features in a database.

3. The sensor device of claim 1 , wherein adjusting the weights of the plurality of machine learning models based on the signature of input features in the CGM sensor data comprises: identify one or more machine learning models of the plurality of machine learning models that are associated with the identified signature of input features; and featuring the one or more machine learning models among the plurality of machine learning models.

4. The sensor device of claim 3 , wherein featuring the one or more machine learning models among the plurality of machine learning models comprises increasing a weighting associated with the one or more machine learning models.

5. The sensor device of claim 3 , wherein featuring the one or more machine learning models among the plurality of machine learning models comprises selecting the one or more machine learning models for generating the output.

6. The sensor device of claim 1 , wherein the training data for each machine learning model is specific to the particular outlier condition.

7. The sensor device of claim 1 , wherein the signature of input features in the CGM sensor data is specific to the particular outlier condition.

8. A method for applying micro machine learning models to reduce sensor glucose signal blanking, the method comprising:
receiving, at a sensor device, CGM sensor data;
inputting the CGM sensor data into a plurality of machine learning models, wherein each machine learning model of the plurality of machine learning models is trained to predict a sensor glucose value under a particular outlier condition using training data comprising clinical data on the particular outlier condition;
identifying a signature of input features in the CGM sensor data;
adjusting weights of the plurality of machine learning models based on the signature of input features in the CGM sensor data;
receiving an output from the plurality of machine learning models, the output indicating a predicted sensor glucose value based on weighting outputs of the plurality of machine learning models using the adjusted weights; and
causing displaying, on a display interface, the output indicating the predicted sensor glucose value such that the sensor glucose signal blanking is reduced.

9. The method of claim 8 , wherein identifying the signature of input features in the CGM sensor data comprises matching a combination of input features in the CGM sensor data with a predetermined signature of input features in a database.

10. The method of claim 8 , wherein adjusting the weights of the plurality of machine learning models based on the signature of input features in the CGM sensor data comprises: identify one or more machine learning models of the plurality of machine learning models that are associated with the identified signature of input features; and featuring the one or more machine learning models among the plurality of machine learning models.

11. The method of claim 10 , wherein featuring the one or more machine learning models among the plurality of machine learning models comprises increasing a weighting associated with the one or more machine learning models.

12. The method of claim 10 , wherein featuring the one or more machine learning models among the plurality of machine learning models comprises selecting the one or more machine learning models for generating the output.

13. The method of claim 8 , wherein the training data for each machine learning model is specific to the particular outlier condition.

14. The method of claim 8 , wherein the signature of input features in the CGM sensor data is specific to the particular outlier condition.

15. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause operations comprising:
receiving, at a sensor device, CGM sensor data;
inputting the CGM sensor data into a plurality of machine learning models, wherein each machine learning model of the plurality of machine learning models is trained to predict a sensor glucose value under a respective, particular outlier condition using training data known to comprise clinical data on the respective, particular outlier condition;
identifying a signature of input features in the CGM sensor data;
adjusting weights of the plurality of machine learning models based on the signature of input features in the CGM sensor data;
receiving an output from the plurality of machine learning models, the output indicating a predicted sensor glucose value based on weighting outputs of the plurality of machine learning models using the adjusted weights; and
causing displaying, on a display interface, the output indicating the predicted sensor glucose value such that the sensor glucose signal blanking is reduced.

16. The non-transitory computer-readable medium of claim 15 , wherein identifying the signature of input features in the CGM sensor data comprises matching a combination of input features in the CGM sensor data with a predetermined signature of input features in a database.

17. The non-transitory computer-readable medium of claim 15 , wherein adjusting the weights of the plurality of machine learning models based on the signature of input features in the CGM sensor data comprises: identify one or more machine learning models of the plurality of machine learning models that are associated with the identified signature of input features; and featuring the one or more machine learning models among the plurality of machine learning models.

18. The sensor device of claim 1 , wherein the particular outlier condition includes at least one of:
an outlier type of user;
an outlier environmental condition;
an outlier sensor wearing condition;
an outlier manufacturing condition; or
an outlier user activity condition.

Assignees

• Medtronic Minimed Inc

Inventors

• Ajemba Peter
• Nogueira Keith

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• A61B5/7475

  User input or interface means, e.g. keyboard, pointing device, joystick · CPC title

• G06N20/20 Primary

  Ensemble learning · CPC title

• G16H50/30

  for calculating health indices; for individual health risk assessment · CPC title