Model mosaic framework for modeling glucose sensitivity
Methods, systems, and devices for modeling a relationship between glucose sensitivity and a sensor electrical property are described herein. More particularly, the methods, systems, and devices describe partitioning an input signal feature space relating glucose sensitivity and a sensor electrical property into subspaces and training a model for each subspace. For example, the subspace models may form a mosaic of models, for which the output is more accurate than a single model.
1. A system 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 using training data comprising clinical data on glucose sensitivity; and
one or more processors configured to:
partition an input signal feature space into a plurality of contiguous subspaces, the input signal feature space relating glucose sensitivity and a sensor electrical property associated with a glucose sensor device, each subspace of the plurality of contiguous subspaces associated with a respective glucose sensitivity behavior of the glucose sensor device that is modeled by a respective machine learning model in the plurality of machine learning models;
receive sensor data relating to the sensor electrical property;
input the sensor data into a machine learning model of the plurality of machine learning models, wherein the machine learning model is selected from the plurality of machine learning models based on the sensor data and a respective range of values associated with the sensor electrical property for each subspace of the plurality of contiguous subspaces;
receive an output from the machine learning model indicating the glucose sensitivity of the glucose sensor device;
determine a glucose level based on the sensor data and the glucose sensitivity; and
control delivery of insulin by an insulin delivery device based on the determined glucose level.
2. The system of claim 1 , wherein the one or more processors are further configured to, based on the output from the machine learning model and one or more criteria:
blank the glucose sensor device when the output from the machine learning model is noncompliant with the one or more criteria; or
utilize readings from the glucose sensor device for glucose level monitoring or glucose level management when the output from the machine learning model is compliant with the one or more criteria.
3. The system of claim 1 , wherein the glucose sensitivity is measured based on an interstitial current signal.
4. The system of claim 1 , wherein the sensor electrical property includes wear time of the glucose sensor device.
5. The system of claim 1 , wherein, to partition the input signal feature space, the one or more processors are further configured to determine the plurality of contiguous subspaces such that the glucose sensitivity, with respect to the sensor electrical property, behaves in a same manner within each subspace of the plurality of contiguous subspaces.
6. The system of claim 1 , wherein the training data for each machine learning model of the plurality of machine learning models is partitioned according to the plurality of contiguous subspaces.
7. The system of claim 1 , wherein the training data is weighted according to the plurality of contiguous subspaces.
8. The system of claim 1 , wherein the one or more processors are further configured to:
obtain a plurality of models including the machine learning model, wherein the plurality of models are ranked from simplest to most complex;
test each model of the plurality of models from simplest to most complex based on one or more criteria; and
determine that the machine learning model is a simplest model that satisfies the one or more criteria.
9. The system of claim 1 , wherein partitioning the input signal feature space into the plurality of contiguous subspaces comprises partitioning the input signal feature space based on ranges of the sensor electrical property, sensor operating conditions, or a combination thereof.
10. The system of claim 9 , wherein the sensor electrical property and sensor operating conditions include wear time, battery life, calibration data, environmental property, a maximum or minimum acceptable glucose measurement, a range of acceptable glucose measurements, a glycemic range, a calibration factor, or a combination thereof.
11. A method comprising, by one or more processors:
partitioning an input signal feature space into a plurality of contiguous subspaces, the input signal feature space relating glucose sensitivity and a sensor electrical property associated with a glucose sensor device, each subspace of the plurality of contiguous subspaces associated with a respective glucose sensitivity behavior of the glucose sensor device that is modeled by a respective machine learning model of a plurality of machine learning models;
receiving sensor data relating to the sensor electrical property;
inputting the sensor data into a machine learning model of the plurality of machine learning models, wherein the machine learning model is selected from the plurality of machine learning models based on the sensor data and a respective range of values associated with the sensor electrical property for each subspace of the plurality of contiguous subspaces, wherein each machine learning model of the plurality of machine learning models is trained using training data comprising clinical data on glucose sensitivity;
receiving an output from the machine learning model indicating the glucose sensitivity of the glucose sensor device;
determining a glucose level based on the sensor data and the glucose sensitivity; and
controlling an insulin delivery device to deliver insulin based on the determined glucose level.
12. The method of claim 11 , further comprising, based on the output from the machine learning model and one or more criteria:
blanking the glucose sensor device when the output from the machine learning model is noncompliant with the one or more criteria; or
utilizing readings from the glucose sensor device for glucose level monitoring or glucose level management when the output from the machine learning model is compliant with the one or more criteria.
13. The method of claim 11 , wherein the glucose sensitivity is measured based on an interstitial current signal (“Isig”).
14. The method of claim 11 , wherein the sensor electrical property includes wear time of the glucose sensor device.
15. The method of claim 11 , further comprising determining the plurality of contiguous subspaces such that the glucose sensitivity, with respect to the sensor electrical property, behaves in a same manner within each subspace of the plurality of contiguous subspaces.
16. The method of claim 11 , wherein the training data for each machine learning model of the plurality of machine learning models is partitioned according to the plurality of contiguous subspaces.
17. The method of claim 11 , wherein the training data is weighted according to the plurality of contiguous subspaces.
18. The method of claim 11 , further comprising:
obtaining a plurality of models including the machine learning model, wherein the plurality of models are ranked from simplest to most complex;
testing each model of the plurality of models from simplest to most complex based on one or more criteria; and
determining that the machine learning model is a simplest model that satisfies the one or more criteria.
19. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause performance of operations comprising:
partitioning an input signal feature space into a plurality of contiguous subspaces, the input signal feature space relating glucose sensitivity and a sensor electrical property associated with a glucose sensor device, each subspace of the plurality of contiguous subspaces associated with a respective glucose sensitivity behavior of the glucose sensor device that is modeled by a respective machine learning model in a plurality of machine learning models;
receiving sensor data relating to the sensor electrical property;
inputting the sensor data into a machine learning model of the plurality of machine learning models, wherein the machine learning model is selected from the plurality of machine learning models based on the sensor data and a respective range of values associated with the sensor electrical property for each subspace of the plurality of contiguous subspaces, wherein each machine learning model of the plurality of machine learning models is trained using training data comprising clinical data on glucose sensitivity;
receiving an output from the machine learning model indicating the glucose sensitivity of the glucose sensor device;
determining a glucose level based on the sensor data and the glucose sensitivity; and
controlling an insulin delivery device to deliver an amount of insulin based on the determined glucose level.
20. The non-transitory computer-readable medium of claim 19 , wherein the operations further comprise, based on the output from the machine learning model and one or more criteria:
blanking the glucose sensor device when the output from the machine learning model is noncompliant with the one or more criteria; or
utilizing readings from the glucose sensor device for glucose level monitoring or glucose level management when the output from the machine learning model is compliant with the one or more criteria.