Method of screening for disorders of glucose metabolism

• US 20040162678A1
• Filed: 11/05/2003
• Published: 08/19/2004
• Est. Priority Date: 08/13/2001
• Status: Abandoned Application

• Current Assignee: Sensys Medical, Inc. (Sensys Medical Limited)
• Original Assignee: Sensys Medical, Inc. (Sensys Medical Limited)
• Inventors: Cone, Andrew G., Ruchti, Timothy L., Hockersmith, Linda, Monfre, Stephen L., Hazen, Kevin H., Blank, Thomas B., Hetzel, Donald
• Application Number:
  US10/702,710
• Publication Number:
  US 20040162678A1
• Time in Patent Office: Days
• US Class Current: 702/19
• CPC Class Codes: A61B 5/14532 for measuring glucose, e.g. by tissue impedance measurement
• 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: C12Q 1/54 involving glucose or galactose
• CPC Class Codes: G01N 2800/042 Disorders of carbohydrate metabolism, e.g. diabetes, glucose metabolism
• CPC Class Codes: G01N 33/66 involving blood sugars, e.g. galactose
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients
• CPC Class Codes: G16Z 99/00 Subject matter not provided for in other main groups of this subclass

Abstract

A method of screening for disorders of glucose metabolism such as impaired glucose tolerance and diabetes allows prevention, or early detection and treatment of diabetic complications such as cardiovascular disease, retinopathy, and other disorders of the major organs and systems. A mathematical algorithm evaluates the shape of a subject'"'"'s glucose profile and classifies the profile into one of several predefined clusters, each cluster corresponding either to a normal condition or one of several abnormal conditions. The series of blood glucose values making up the glucose tolerance curve may be measured using any glucose analyzer including: invasive, minimally invasive and noninvasive types. The method is executed on a processing device programmed to perform the steps of the method. Depending on the outcome of the screening, a subject may be provided with additional information concerning their condition and/or counseled to consult further with their health care provider.

Claims

1. A method of screening a subject for disorders of glucose metabolism, comprising steps of:

measuring at least a portion of a glucose profile, said profile comprising a plurality of blood glucose values from at least after a glucose challenge;

extracting features from said at least a portion of said profile, wherein features comprise characteristics of said at least a portion of said profile relevant for classification; and

classifying said subject on the basis of said features.

2. The method of claim 1, further comprising a step of processing said at least a portion of said glucose profile, wherein at least one transformation is applied to eliminate or attenuate interference and to correct said at least a portion of said profile, so that a signal of interest is enhanced and made accessible for analysis.

3. The method of claim 2, wherein said at least one transformation includes any of:

detection of outliers through statistical and model based methods that exploit the properties of the profile;

autocorrelation;

non-causal filtering of the profile;

time series analysis and optimum filtering techniques;

phase and magnitude correction related to known error distributions between a measured profile and reference glucose measurements;

mean-centering;

baseline correction;

normalization;

multivariate signal correction;

standard normal variate transformation;

calculating one or both of first and second derivatives of the profile; and

state transformations.

4. The method of claim 2, wherein a processed measurement, y∈

^N, is determined according to

y=h(t,x),

where h;

<sup>N×

2</sup>→

^N is the preprocessing function, x∈

^N is the glucose measurements and t∈

^N is the vector of times associated with each glucose measurement.

5. The method of claim 2, wherein said step of extracting features comprises decomposing processed data into abstract features, wherein abstract features comprise any of:

principal components;

wavelet basis components; and

Fourier coefficients.

6. The method of claim 2, wherein said step of processing said at least a portion of said glucose profile comprises enhancing said least a portion of said glucose profile through any of:

outlier analysis;

filtering; and

magnitude and/or phase correction;

prior to analysis by a healthcare provider.

7. The method of claim 2, wherein said step of processing said at least a portion of said glucose profile comprises calculating any of first and second derivatives of said least a portion of said glucose profile.

8. The method of claim 1, wherein feature extraction comprises any mathematical transformation that enhances a quality or aspect of the profile for interpretation or classification.

9. The method of claim 8, wherein feature extraction concisely represents the information content of said profile in the simplest and most accessible form prior to application of a classification algorithm, so that the greatest discrimination between various classes is provided.

10. The method of claim 2, wherein said step of extracting features comprises a step of representing said features in a vector, z∈

^M that is determined from the processed profile through

z=f(t,y)

wherein f;

<sup>N×

2</sup>→

^M is a mapping from a measurement space to a feature space.

11. The method of claim 10, wherein decomposing f(•

) yields specific transformations, f₁(•

);

^N→

^M_i for determining a specific feature; and

wherein a dimension, M_i, indicates whether an i^th feature is a scalar or a vector and the aggregation of all features is the vector z.

12. The method of claim 11, wherein a feature that is represented as a vector or a pattern exhibits a structure indicative of an underlying physical phenomenon.

13. The method of claim 11, wherein features are either abstract or simple features.

14. The method of claim 13, wherein abstract features do not necessarily have a specific interpretation related to the physical system.

15. The method of claim 14, wherein abstract features comprises scores of a principal component analysis.

16. The method of claim 13, wherein simple features can be related directly to said processed profile.

17. The method of claim 16, wherein said simple features include any of:

first and second derivative at key time points; and

duration between various time points.

18. The method of claim 13, wherein compilation of abstract and simple features constitutes the M-dimensional feature space.

19. The method of claim 13, wherein optimum feature selection and/or data compression is applied to enhance the robustness of a classifier, due to redundancy of information across the set of features.

20. The method of claim 2, wherein features further comprise known information unrelated to said profile.

21. The method of claim 20, wherein said known information unrelated to said profile includes any of:

age;

history of diabetes;

weight;

height;

body mass index;

gender;

ethnicity;

diet and/or exercise patterns;

HbA1c level; and

insulin and/or c-peptide level.

22. The method of claim 1, wherein said step of classifying said subject on the basis of said features comprises:

defining classes;

mapping said features to said classes;

and assigning class membership by a decision engine;

wherein a subject classification related to a particular disorder of glucose metabolism is determined.

23. The method of claim 22, wherein said classes correspond either to a normal state or to one of a plurality of disorders of glucose metabolism.

24. The method of claim 22, wherein said step of defining classes comprises:

assigning measurements from an exploratory data set to classes, said data set comprising exemplar features from a representative sampling of a subject population.

25. The method of claim 24, wherein classes are defined in any of a supervised and an unsupervised manner.

26. The method of claim 25, wherein the step of defining classes in a supervised manner comprises classes defining classes through known differences in the data, wherein use of a priori information develops classification models when class assignment is known.

27. The method of claim 25, wherein the step of defining classes in an unsupervised manner comprises using the exemplar features to develop clusters or natural groupings of the data in the feature space;

wherein within cluster homogeneity and between cluster separation are optimized, and wherein clusters formed from features with physical meaning are interpreted based on known underlying phenomenon causing variation in the feature space.

28. The method of claim 25, wherein supervised and unsupervised approaches are combined to utilize a priori knowledge and exploration of the feature space for naturally occurring spectral classes.

29. The method of claim 25, further comprising steps of:

dividing each set of features into a plurality of regions; and

defining classes by combinations of said regions;

wherein classes are defined from features in a supervised manner.

30. The method of claim 29, further comprising steps of:

performing cluster analysis on the data;

comparing results of said cluster analysis with said classes defined from features in a supervised manner; and

using clusters to determine groups of classes that can be combined, wherein the number of final class definitions is reduced according to natural divisions in the data.

31. The method of claim 30, further comprising a step of designing a classifier based on supervised pattern recognition.

32. The method of claim 31, wherein said step of designing said classifier based on supervised pattern recognition comprises steps of:

creating a model based on class definitions that transforms a measured set of features to an estimated classification; and

optimizing class definitions using an iterative approach to satisfy specifications of a measurement system, wherein said classifier produces a robust and accurate subject assessment.

33. The method of claim 32, wherein said classes are mutually exclusive and wherein said step of mapping said features to said classes comprises assigning each measurement to one class.

34. The method of claim 33, wherein variation of said mutually exclusive classes is described statistically through application of statistical classification methods.

35. The method of claim 33, wherein said step of designing a classifier comprises determining an optimal mapping or transformation from the feature space to a class estimate that minimizes the number of misclassification.

36. The method of claim 35, wherein said mapping is based on any of:

linear discriminant analysis;

SIMCA (soft independent modeling of class analogies);

k nearest-neighbor; and

artificial neural networks.

37. The method of claim 35, wherein said classifier comprises either a function or an algorithm that maps the feature to a class, c, according to:

c=f(z),

where c is an integer on an interval [1,P] and P is the number of classes.

38. The method of claim 32, wherein a fuzzy classification allows class membership in more than one class simultaneously.

39. The method of claim 38, further comprising a step of providing a measure relating to the extent to which a particular feature set is related to a given class.

40. The method of claim 38, wherein membership in fuzzy sets is defined by a continuum of grades and a set of membership functions that map the feature space into an interval [0,1] for each class.

41. The method of claim 40, wherein an assigned membership grade represents the degree of class membership, wherein a value of 1 corresponds to the highest degree;

wherein a sample can simultaneously be a member of more than one class.

42. The method of claim 38, wherein mapping from feature space to a vector of class memberships is given by

c_k=f_k(z),

where k=1, 2, . . . P, f_k(•

) is the membership function of the k^th class, c_k∈

[0,1] for all k and the vector c∈

^P is the set of class memberships;

wherein the membership vector provides the degree of membership in each of the predefined classes.

43. The method of claim 1, wherein said step of measuring at least a portion of a glucose profile comprises measuring at least a portion of a glucose profile, said profile comprising a plurality of blood glucose values from before and after a glucose challenge;

44. A method of classifying a subject based on a glucose profile, comprising steps of:

extracting features from at least a portion of said glucose profile, said features comprising characteristics of said at least a portion of said profile relevant for classification;

defining classes;

mapping said features for classification; and

assigning class membership;

wherein a subject classification related to a particular disorder of glucose metabolism is determined.

45. The method of claim 44, wherein said classes correspond either to a normal state or to one of a plurality of disorders of glucose metabolism.

46. The method of claim 44, wherein said step of defining classes comprises:

assigning measurements from an exploratory data set to classes, said data set comprising exemplar features from a representative sampling of a subject population.

47. The method of claim 46, wherein classes are defined in any of a supervised and an unsupervised manner.

48. The method of claim 47, wherein the step of defining classes in a supervised manner comprises classes defining classes through known differences in the data, wherein use of a priori information develops classification models when class assignment is known.

49. The method of claim 47, wherein the step of defining classes in an unsupervised manner comprises using the exemplar features to develop clusters or natural groupings of the data in the feature space;

wherein within cluster homogeneity and between cluster separation are optimized, and wherein clusters formed from features with physical meaning are interpreted based on known underlying phenomenon causing variation in the feature space.

50. The method of claim 47, wherein supervised and unsupervised approaches are combined to utilize a priori knowledge and exploration of the feature space for naturally occurring spectral classes.

51. The method of claim 47, further comprising steps of:

dividing each set of features into a plurality of regions; and

defining classes by combinations of said regions;

wherein classes are defined from features in a supervised manner.

52. The method of claim 51, further comprising steps of:

performing cluster analysis on the data;

comparing results of said cluster analysis with said classes defined from features in a supervised manner; and

using clusters to determine groups of classes that can be combined, wherein the number of final class definitions is reduced according to natural divisions in the data.

53. The method of claim 52, further comprising a step of designing a classifier based on supervised pattern recognition.

54. The method of claim 53, wherein said step of designing said classifier based on supervised pattern recognition comprises steps of:

creating a model based on class definitions that transforms a measured set of features to an estimated classification; and

optimizing class definitions using an iterative approach to satisfy specifications of a measurement system, wherein said classifier produces a robust and accurate subject assessment.

55. The method of claim 54, wherein said classes are mutually exclusive and wherein said step of mapping said features to said classes comprises assigning each measurement to one class.

56. The method of claim 55, wherein variation of said mutually exclusive classes is described statistically through application of statistical classification methods.

57. The method of claim 55, wherein said step of designing a classifier comprises determining an optimal mapping or transformation from the feature space to a class estimate that minimizes the number of misclassification.

58. The method of claim 57, wherein said mapping is based on any of:

linear discriminant analysis;

SIMCA (soft independent modeling of class analogies);

k nearest-neighbor; and

artificial neural networks.

59. The method of claim 57, wherein said classifier comprises either a function or an algorithm that maps the feature to a class, c, according to:

c=f(z),

where c is an integer on an interval [1,P] and P is the number of classes.

60. The method of claim 54, wherein a fuzzy classification allows class membership in more than one class simultaneously.

61. The method of claim 60, further comprising a step of providing a measure relating to the extent to which a particular feature set is related to a given class.

62. The method of claim 60, wherein membership in fuzzy sets is defined by a continuum of grades and a set of membership functions that map the feature space into an interval [0,1] for each class.

63. The method of claim 62, wherein an assigned membership grade represents the degree of class membership, wherein a value of 1 corresponds to the highest degree;

wherein a sample can simultaneously be a member of more than one class.

64. The method of claim 60, wherein mapping from feature space to a vector of class memberships is given by

c_k=f_k(z),

where k=1,2, . . . P, f_k(•

) is the membership function of the k^th class, c_k∈

[0,1] for all k and the vector c∈

^P is the set of class memberships;

wherein the membership vector provides the degree of membership in each of the predefined classes.

65. The method of claim 44, further comprising a step of processing said at least a portion of said glucose profile, wherein at least one transformation is applied to eliminate or attenuate interference and to correct said at least a portion of said profile, so that a signal of interest is enhanced and made accessible for analysis.