Sensor calibration using simulated sensor measurements

Abstract

Techniques for sensor calibration involve determining a generative model using sensor measurements from different instances of a first glucose sensor together with corresponding reference glucose values. The generative model is configured to generate a simulated measurement representing a predicted output of the first glucose sensor under specific operating conditions. A set of simulated measurements is generated using operating conditions observed with respect to a second glucose sensor as inputs to the generative model. The second glucose sensor is a sensor of a different design. The operating conditions observed with respect to the second glucose sensor include reference glucose values obtained in connection with measurements made using the second glucose sensor. The simulated measurements are then used to determine an estimation model for the first glucose sensor. The estimation model is configured to estimate glucose level given one or more sensor measurements from the first glucose sensor.

What is claimed is: 
     
         1 . A method of calibrating a first glucose sensor, the method comprising:
 obtaining sensor measurements from different instances of the first glucose sensor together with corresponding reference glucose values;   determining, by one or more processors of a computer system, a generative model for the first glucose sensor, wherein:
 the generative model is configured to generate a simulated measurement representing a predicted output of the first glucose sensor under specific operating conditions, and 
 determining the generative model comprises identifying relationships between the sensor measurements and the reference glucose values; 
   generating a set of simulated measurements using the generative model, wherein:
 generating the set of simulated measurements comprises applying operating conditions observed with respect to a second glucose sensor as inputs to the generative model, 
 the second glucose sensor has a different design than the first glucose sensor, and 
 the operating conditions observed with respect to the second glucose sensor include reference glucose values obtained in connection with measurements made using the second glucose sensor; 
   determining, by the one or more processors, an estimation model based on relationships between the simulated measurements and the reference glucose values obtained in connection with measurements made using the second glucose sensor, wherein the estimation model is configured to estimate glucose level given one or more sensor measurements from the first glucose sensor; and   communicating the estimation model to an electronic device that applies the estimation model to sensor measurements from a particular instance of the first glucose sensor.   
     
     
         2 . The method of  claim 1 , wherein the sensor measurements from different instances of the first glucose sensor comprise parameters characterizing an electrochemical response of the first glucose sensor. 
     
     
         3 . The method of  claim 1 , wherein the first glucose sensor and the second glucose sensor are interstitial glucose sensors. 
     
     
         4 . The method of  claim 1 , wherein the reference glucose values corresponding to the sensor measurements from different instances of the first glucose sensor and the reference glucose values obtained in connection with measurements made using the second glucose sensor comprise blood glucose values from a blood glucose meter. 
     
     
         5 . The method of  claim 1 , further comprising:
 determining a delivery command for an insulin infusion device based on a glucose level resulting from applying the estimation model to the sensor measurements from the particular instance of the first glucose sensor.   
     
     
         6 . The method of  claim 1 , wherein the operating conditions observed with respect to the second glucose sensor are represented by different sets of input variables to the generative model, each set of input variables being associated with a corresponding instance of the second glucose sensor and comprising a reference glucose value in combination with:
 a calibration factor specific to the corresponding instance of the second glucose sensor,   an age of the corresponding instance of the second glucose sensor, or   both the calibration factor and the age of the corresponding instance of the second glucose sensor.   
     
     
         7 . The method of  claim 6 , wherein each set of input variables to the generative model includes the calibration factor specific to the corresponding instance of the second glucose sensor, and wherein each simulated measurement represents a predicted output of the first glucose sensor for the calibration factor indicated in one of the sets of input variables. 
     
     
         8 . The method of  claim 6 , wherein each set of input variables to the generative model includes the age of the corresponding instance of the second glucose sensor, and wherein each simulated measurement represents a predicted output of the first glucose sensor at the age indicated in one of the sets of input variables. 
     
     
         9 . The method of  claim 6 , wherein determining the generative model comprises:
 training a first neural network to generate the simulated measurements, using calibration factor values, age values, and reference glucose values as inputs to the first neural network, and using corresponding sensor measurements as outputs to be produced by the first neural network.   
     
     
         10 . The method of  claim 9 , further comprising:
 training a second neural network to generate an indication of whether an output of the first neural network is plausible given a particular combination of inputs to the first neural network, wherein the second neural network is trained using the same sensor measurements as the first neural network.   
     
     
         11 . The method of  claim 10 , further comprising:
 training the first neural network in concert with the second neural network such that the first neural network generates simulated measurements that are optimized based on output of the second neural network.   
     
     
         12 . A computer system comprising:
 one or more processors; and   one or more processor-readable media storing instructions which, when executed by the one or more processors, cause performance of:
 obtaining sensor measurements from different instances of a first glucose sensor together with corresponding reference glucose values; 
 determining a generative model for the first glucose sensor, wherein:
 the generative model is configured to generate a simulated measurement representing a predicted output of the first glucose sensor under specific operating conditions, and 
 determining the generative model comprises identifying relationships between the sensor measurements and the reference glucose values; 
 
 generating a set of simulated measurements using the generative model, wherein:
 generating the set of simulated measurements comprises applying operating conditions observed with respect to a second glucose sensor as inputs to the generative model, 
 the second glucose sensor has a different design than the first glucose sensor, and 
 the operating conditions observed with respect to the second glucose sensor include reference glucose values obtained in connection with measurements made using the second glucose sensor; 
 
 determining an estimation model based on relationships between the simulated measurements and the reference glucose values obtained in connection with measurements made using the second glucose sensor, wherein the estimation model is configured to estimate glucose level given one or more sensor measurements from the first glucose sensor; and 
 communicating the estimation model to an electronic device that applies the estimation model to sensor measurements from a particular instance of the first glucose sensor. 
   
     
     
         13 . The computer system of  claim 12 , wherein the sensor measurements from different instances of the first glucose sensor comprise parameters characterizing an electrochemical response of the first glucose sensor. 
     
     
         14 . The computer system of  claim 12 , wherein the first glucose sensor and the second glucose sensor are interstitial glucose sensors. 
     
     
         15 . The computer system of  claim 12 , wherein the reference glucose values corresponding to the sensor measurements from different instances of the first glucose sensor and the reference glucose values obtained in connection with measurements made using the second glucose sensor comprise blood glucose values from a blood glucose meter. 
     
     
         16 . The computer system of  claim 12 , wherein the operating conditions observed with respect to the second glucose sensor are represented by different sets of input variables to the generative model, each set of input variables being associated with a corresponding instance of the second glucose sensor and comprising a reference glucose value in combination with:
 a calibration factor specific to the corresponding instance of the second glucose sensor,   an age of the corresponding instance of the second glucose sensor, or   both the calibration factor and the age of the corresponding instance of the second glucose sensor.   
     
     
         17 . The computer system of  claim 16 , wherein each set of input variables to the generative model includes the calibration factor specific to the corresponding instance of the second glucose sensor, and wherein each simulated measurement represents a predicted output of the first glucose sensor for the calibration factor indicated in one of the sets of input variables. 
     
     
         18 . The computer system of  claim 16 , wherein each set of input variables to the generative model includes the age of the corresponding instance of the second glucose sensor, and wherein each simulated measurement represents a predicted output of the first glucose sensor at the age indicated in one of the sets of input variables. 
     
     
         19 . The computer system of  claim 16 , wherein determining the generative model comprises:
 training a first neural network to generate the simulated measurements, using calibration factor values, age values, and reference glucose values as inputs to the first neural network, and using corresponding sensor measurements as outputs to be produced by the first neural network.   
     
     
         20 . The computer system of  claim 19 , wherein when executed by the one or more processors, the instructions further cause performance of:
 training a second neural network to generate an indication of whether an output of the first neural network is plausible given a particular combination of inputs to the first neural network, wherein the second neural network is trained using the same sensor measurements as the first neural network.