GLUCOSE PREDICTOR BASED ON REGULARIZATION NETWORKS WITH ADAPTIVELY CHOSEN KERNELS AND REGULARIZATION PARAMETERS

• US 20140073892A1
• Filed: 04/20/2012
• Published: 03/13/2014
• Est. Priority Date: 04/20/2011
• Status: Active Grant

• Current Assignee: Novo Nordisk A/S
• Original Assignee: Novo Nordisk A/S
• Inventors: Randloev, Jette, McKennoch, Samuel, Pereverzyev, Sergei, Sampath, Sivananthan
• Granted Patent: US 10,307,109 B2
• Time in Patent Office: Days
• US Class Current: 600/365
• CPC Class Codes: A61B 5/14532 for measuring glucose, e.g. by tissue impedance measurement
• CPC Class Codes: A61B 5/4848 Monitoring or testing the effects of treatment, e.g. of medication
• CPC Class Codes: A61B 5/4866 Evaluating metabolism A61B5/083 takes precedence
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: A61B 5/7275 Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
• CPC Class Codes: G16H 50/50 for simulation or modelling of medical disorders

Abstract

The invention relates to a method and a device for predicting a glycaemic profile of a subject. A multistage algorithm is employed comprising a prediction setting stage specifying a functional space for the prediction and a prediction execution stage specifying a predicted future glycaemic state of the subject in the functional space as a continuous function of time.

Claims

1. A glucose prediction device comprising:

Input device structured to receive information indicative of a physiologic condition of a subject,

processing device structured to predict a glycaemic state of the subject on the basis of information received by the input device, and

output device structured to convey information related to the predicted glycaemic state,

wherein the processing device is adapted to perform a multistage prediction process comprising a prediction setting stage specifying a functional space for the prediction and a prediction execution stage specifying a predicted glycaemic state of the subject in the functional space as a continuous function of time.

2. A device according to claim 1, wherein the prediction setting stage specifies the functional space based on the information received by the input device.

3. A device according to claim 1, wherein the specified continuous function of time depends on a reproducing kernel and a regularization parameter, and wherein the prediction setting stage is adapted to determine the reproducing kernel and an initial value of the regularisation parameter.

4. A device according to claim 1, wherein the prediction setting stage comprises:

a labelling of data segments containing information received by the input device, and

a determination of a kernel and an initial regularization parameter based on respective labels of the data segments, and wherein the prediction execution stage specifies a predicted glycaemic state by using the determined kernel and initial regularization parameter to construct the continuous function of time.

5. A device according to claim 4, wherein the labelling of data segments comprises a linear fitting, and wherein the suggestion of a kernel and an initial regularization parameter is based on respective coefficients of the linear fit.

6. A device according to claim 4, wherein the continuous function of time is constructed by a regularized learning algorithm in a reproducing kernel Hilbert space.

7. A device according to claim 6, wherein the reproducing kernel Hilbert space is selected based on a data pool of training data segments, each training data segment being indicative of a physiologic condition of a subject at respective points in time.

8. A device according to claim 6, wherein the prediction setting stage specifies a functional space for the prediction on the basis of information obtained from a predetermined data pool.

9. A device according to claim 1, wherein the prediction setting stage specifies a functional space for the prediction on the basis of information obtained from a continuously updated data pool.

10. A device according to claim 1, wherein the information indicative of a physiologic condition of a subject comprises information related to at least one measurement of a body characteristic.

11. A device according to claim 10, wherein the body characteristic is blood or tissue glucose.

12. A device according to claim 10, wherein the input device is further adapted to receive information related to a therapeutic treatment.

13. A device according to claim 12, wherein the information related to a therapeutic treatment comprises information related to past delivery of a glucose regulating agent.

14. A device according to claim 12, wherein the input device is further adapted to receive information related to a meal consumed or to be consumed by the subject.

20. A computer-implemented method of constructing a prediction process for a glucose prediction device as defined in claim 1, the prediction process comprising a prediction setting stage specifying a functional space for the prediction and a prediction execution stage specifying a predicted glycemic state of a subject in the functional space as a continuous function of time;

the functional space being parameterised by a first set of parameters, the prediction setting stage specifying the first set of parameters as respective functions of an input data segment, each of the respective functions being specified by a respective kernel;

the method comprising;

receiving data pool of training data segments, each training data segment being indicative of a physiologic condition of a subject at respective points in time;

determining the set of respective kernels from the obtained data pool.

21. A method according to claim 20, wherein determining the set of respective kernels from the obtained data pool comprises:

compressing each training data segment to obtain a corresponding set of compressed training data segments;

determining a set of training parameters parameterising said functional space, each training parameter of the training set minimizing a prediction error function indicative of a deviation between a physiologic condition and a predicted physiologic condition, the predicted physiologic condition being predicted based on the functional space parameterised by said training parameter;

constructing, from the set of training parameters, a non-linear mapping between compressed input data segments and a set of kernel parameters parameterising the respective kernels.

22. A method according to claim 21, wherein constructing the non-linear mapping comprises determining a mechanism for determining a Tikhonov regularisation parameter.

23. A method according to claim 21, wherein the prediction error function penalises a delay or failure in the prediction of physiological events classified as dangerous more heavily than a delay or failure in the prediction of physiological events classified as normal.

15. A computer-implemented method for predicting a glucose profile of a subject, the method comprising:

receiving, by a processing device, information related to at least one measurement of a body characteristic,

specifying, by the processing device, a functional space for the prediction on the basis of the information received, and

specifying, by the processing device, a predicted glycemic state of the subject in the functional space as a continuous function of time.

16. A method according to claim 15, wherein specifying a functional space for the prediction comprises labelling data segments containing information received by the input device and determining a kernel and an initial regularization parameter based on respective data segment labels.

17. A method according to claim 16, wherein the kernel is parameterised by a set of kernel parameters;

and wherein determining the kernel and the initial regularization parameter based on respective data segment labels comprises determining the set of kernel parameters based on respective pre-computed predictor functions for predicting kernel parameters as a function of input data segment labels.

18. A method according to claim 17, wherein the pre-computed predictor functions define a reproducing kernel Hilbert space, and wherein determining the kernel parameters comprises computing the kernel parameters by a regularized learning algorithm in the reproducing kernel Hilbert space defined by at least approximately minimizing an error function.

19. A method according to claim 15, wherein specifying a predicted glycemic state of the subject in the functional space comprises using the determined kernel and initial regularization parameter to construct a predictor.

25. A computer program product comprising computer program code adapted to cause a data processing system to perform the method according to claim 15, when said computer program code is executed on said data processing system.

26. A data processing system having stored thereon computer program code adapted to cause the data processing system to perform the steps of the method according to claim 15, when said computer program code is executed on said data processing system.

24. A computer-implemented method for determining a glucose concentration, the method comprising:

receiving a sensor signal of a subject,

estimating a glucose concentration from the received sensor signal;

wherein estimating comprises determining an estimated glucose concentration predictor function in a predetermined functional space as a continuous function of sensor signal values.