Multi-tier method of classifying sample spectra for non-invasive blood analyte prediction

• US 6,512,936 B1
• Filed: 09/18/2000
• Issued: 01/28/2003
• Est. Priority Date: 07/22/1999
• Status: Expired due to Term

• Current Assignee: GLT Acquisition Corporation
• Original Assignee: Sensys Medical, Inc. (Sensys Medical Limited)
• Inventors: Ruchti, Timothy L., Monfre, Stephen L., Blank, Thomas B., Thennadil, Suresh
• Primary Examiner(s): WINAKUR, ERIC FRANK
• Application Number:
  US09/665,201
• Time in Patent Office: 862 Days
• Field of Search: 600/310, 600/322, 600/323, 600/330, 600/331, 600/336, 600/473, 128/920, 128/924, 128/925, 356/402, 250/340, 250/341.1, 250/339.01
• US Class Current: 600/322
• CPC Class Codes: A61B 2560/0223 of calibration, e.g. protocols for calibrating sensors
• CPC Class Codes: A61B 2560/0233 Optical standards
• CPC Class Codes: A61B 5/1075 for measuring dimensions by non-invasive methods, e.g. for determining thickness of tissue layer A61B8/0858 takes precedence
• CPC Class Codes: A61B 5/14532 for measuring glucose, e.g. by tissue impedance measurement
• CPC Class Codes: A61B 5/1455 using optical sensors, e.g. spectral photometrical oximeters
• CPC Class Codes: A61B 5/1495 Calibrating or testing of in-vivo probes
• 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: G01N 21/274 Calibration, base line adjustment, drift correction
• CPC Class Codes: G01N 21/359 using near infrared light
• CPC Class Codes: G01N 21/4785 Standardising light scatter apparatus; Standards therefor
• CPC Class Codes: G01N 21/49 within a body or fluid
• CPC Class Codes: G06F 18/21 Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: Y10S 128/92 Computer assisted medical diagnostics

Abstract

A method of multi-tier classification and calibration in noninvasive blood analyte prediction minimizes prediction error by limiting co-varying spectral interferents. Tissue samples are categorized based on subject demographic and instrumental skin measurements, including in vivo near-IR spectral measurements. A multi-tier intelligent pattern classification sequence organizes spectral data into clusters having a high degree of internal consistency in tissue properties. In each tier, categories are successively refined using subject demographics, spectral measurement information and other device measurements suitable for developing tissue classifications. The multi-tier classification approach to calibration utilizes multivariate statistical arguments and multi-tiered classification using spectral features. Variables used in the multi-tiered classification can be skin surface hydration, skin surface temperature, tissue volume hydration, and an assessment of relative optical thickness of the dermis by the near-IR fat band. All tissue parameters are evaluated using the NIR spectrum signal along key wavelength segments.

Claims

1. A multi-tier pattern classification method for estimating a level of a target blood analyte comprising the steps of:

providing a measured tissue absorbance spectrum from a subject;

initially, classifying said measured spectrum into previously defined classes based on a priori information pertaining to said subject;

further classifying said measured spectrum into previously defined classes based on at least one instrumental measurement at a tissue measurement site at which an optical sample was taken for said tissue absorbance spectrum; and

extracting at least one feature from said measured spectrum for still further classification.

2. The pattern classification method of claim 1, wherein said initial classification step comprises the steps of:

in a first tier, classifying said measured spectrum into previously defined classes based on subject'"'"'s age; and

in a second tier, further classifying said measured spectrum into previously defined classes based on subject'"'"'s sex.

3. The pattern classification method of claim 1, wherein said further classification step further comprises the steps of;

in a third tier, further classifying said measured spectrum into previously defined classes based on an estimation of stratum corneum hydration at said tissue measurement site; and

in a fourth tier, further classifying said measured spectrum into previously defined classes based on skin temperature at said tissue measurement site.

4. The pattern classification method of claim 3, wherein said stratum corneum hydration estimate is based on a measurement of ambient humidity at said tissue measurement site.

5. The pattern classification method of claim 1, wherein said feature extraction step comprises any mathematical transformation that enhances a quality or aspect of sample measurement for interpretation to represent concisely structural properties and physiological state of a tissue measurement site, wherein a resulting set of features is used to classify a subject and determine a calibration model that is most useful for blood analyte prediction.

6. The pattern classification method of claim 5, wherein said features are represented in a vector, zε

R^M that is determined from a preprocessed measurement through;

7. The pattern classification method of claim 6, wherein individual features are divided into two categories comprising:

abstract features that do not necessarily have a specific interpretation related to a physical system; and

simple features that are derived from an a priori understanding of a sample and that can be related directly to a physical phenomenon.

8. The pattern classification method of claim 7, wherein said simple features can be calculated from NIR spectral absorbance measurements, said simple features including any of:

thickness of adipose tissue;

hematocrit level;

tissue hydration;

magnitude of protein absorbance;

scattering properties of said tissue;

skin thickness;

temperature related effects;

age related effects;

spectral characteristics;

pathlength estimates;

volume fraction of blood in tissue; and

spectral characteristics related to environmental influences.

9. The pattern classification method of claim 1, further comprising the step of:

employing spectral decomposition to determine features related to a known spectral absorbance pattern.

10. The pattern classification method of claim 1, further comprising the step of:

employing factor-based methods to build a model capable of representing variation in a measured absorbance related to a demographic variable;

wherein projection of a measured absorption onto said model constitutes a feature that represents spectral variation related to said demographic variable.

11. The pattern classification method of claim 1, wherein said feature extraction step determines at least one calibration model that is most appropriate for measurement;

wherein a subject is assigned to one of many predefined classes for which a calibration model has been developed and tested.

12. The pattern classification method of claim 1, further comprising the steps of;

measuring the similarity of a feature to predefined classes; and

assigning class membership.

13. The pattern classification method of claim 12, wherein said assigning step uses mutually exclusive classes and assigns each measurement to one class.

14. The pattern classification method of claim 12, wherein said assigning step uses a fuzzy classification that allows membership in more than one class simultaneously and provides a number between zero and one indicating a degree of membership in each class.

15. The pattern classification method of claim 1, further comprising the step of:

assigning measurements in an exploratory data set to classes.

16. The pattern classification method of claim 15, further comprising the step of;

using measurements and class assignments to determine a mapping from features to class assignments.

17. The pattern classification method of claim 16, further comprising the steps of:

defining classes from said features in a supervised manner, wherein each set of features is divided into two or more regions, and wherein classes are defined by combination of feature divisions;

performing a cluster analysis on the spectral data to determine groups of said defined classes that can be combined, wherein the final number of class definitions is significantly reduced;

designing a classifier subsequent to class definition through supervised pattern recognition by determining an optimal mapping or transformation from the feature space to a class estimate that minimizes the number of misclassifications; and

creating a model based on class definitions that transforms a measured set of features to an estimated classification, wherein said class definitions are optimized to satisfy the specifications of the measurement system.

18. The pattern classification method of claim 17, wherein said optimal mapping utilizes any of linear discriminant analysis, SIMCA, k nearest-neighbor, and artificial neural networks.

19. The pattern classification method of claim 18, wherein a classification function maps said feature to a class c, according to

20. The pattern classification method of claim 19, further comprising the step of:

passing said classification to a nonlinear model that provides a blood analyte prediction based on said classification and spectral measurement, wherein said blood analyte prediction for a measurement x is given by;

21. The pattern classification method of claim 20, wherein a different calibration is realized for each class and wherein an estimate of blood analyte concentration for a measurement is given by:

22. The pattern classification method of claim 17, wherein a membership function maps said feature space into an interval [0,1] for each class, wherein membership is defined by a continuum of grades, and wherein a mapping from feature space to a vector of class memberships is given by:

23. The pattern classification method of claim 22, further comprising the step of predicting a blood analyte by application of a calibration model to a preprocessed measurement.

24. The pattern classification method of claim 23, wherein said calibration model comprises any of nonlinear regression, nonlinear partial least squares, and artificial neural networks.

25. The pattern classification method of claim 24, wherein the calibration model is passed a vector of class memberships, where a vector, c, is used to determine an adaptation of said calibration model suitable for blood analyte prediction or an optimal combination of several blood analyte predictions.

26. The pattern classification method of claim 25, wherein a membership vector and preprocessed absorbance spectrum are both used by the calibration model for blood analyte prediction where the calculation is given by:

27. The pattern classification method of claim 26, wherein separate calibrations are used for each class;

and wherein each calibration is generated using all measurements in a calibration set by exploiting a membership vector assigned to each measurement.

28. The pattern classification method of claim 27, wherein said membership vector is used to determine an optimal combination of p blood analyte predictions from all classes through defuzzification.

29. The pattern classification method of claim 28, wherein each of the p calibration models is developed using an entire calibration set.

30. The pattern classification method of claim 29, wherein calibration measurements are weighted by their respective membership in a k^th class when a k^th calibration model is calculated, where weighted least squares is applied to calculated regression coefficients in a linear case, and wherein a covariance matrix is used in a factor-based methods case.

31. The pattern classification method of claim 30, wherein said defuzzification is a mapping from a vector of blood analyte predictions and a vector of class memberships to a single analyte prediction, wherein said defuzzifier can be denoted as a transformation such that:

 

ŷ

=d(c,└

y₁y₂y₃ . . . y_p┘

),

where d(•

) is the defuzzification function, c is a class membership vector and y_k is a blood analyte prediction of a k^th calibration model.

32. A pattern classification method for estimating a level of a target blood analyte comprising the steps of:

providing a measured tissue absorbance spectrum from a subject;

in at least one tier, classifying said measured spectrum into previously defined classes; and

extracting at least one feature from said measured spectrum for still further classification.

33. The pattern classification method of claim 32, wherein said classifying step is based on any of:

abstract and simple features.

34. The pattern classification method of claim 32, further comprising the step of mapping said measured spectrum to an estimate of said analyte based on either a linear or a nonlinear model.

35. The pattern classification method of claim 32, wherein said classifying step is based on any of:

a priori information; and

at least one instrumental measurement at a tissue measurement site at which an optical sample was taken for said tissue absorbance spectrum.

36. The pattern classification method of claim 32, wherein said classifying step comprises multiple tiers.

37. The pattern classification method of claim 36, wherein said classifying step comprises any of the steps of:

classifying said measured spectrum into previously defined classes based on subject'"'"'s age;

classifying said measured spectrum into previously defined classes based on subject'"'"'s sex;

classifying said measured spectrum into previously defined classes based on an estimation of stratum corneum hydration at said tissue measurement site; and

classifying said measured spectrum into previously defined classes based on skin temperature at said tissue measurement site.

38. A pattern classification method for estimating a level of a target blood analyte comprising the steps of:

providing a measured tissue absorbance spectrum from a subject;

in at least one tier, classifying said measured spectrum into previously defined classes;

extracting at least one feature from said measured spectrum for still further classification; and

estimating said blood analyte through application of a calibration model to said measured spectrum.

39. The pattern classification method of claim 38, wherein said classifying step is based on any of:

abstract and simple features.

40. The pattern classification method of claim 38, wherein said model is either linear or nonlinear.

41. The pattern classification method of claim 38, wherein said classifying step is based on any of:

a priori information; and

at least one instrumental measurement at a tissue measurement site at which an optical sample was taken for said tissue absorbance spectrum.

42. The pattern classification method of claim 38, wherein said classifying step comprises multiple tiers.

43. The pattern classification method of claim 42, wherein said classifying step comprises any of the steps of:

classifying said measured spectrum into previously defined classes based on subject'"'"'s age;

classifying said measured spectrum into previously defined classes based on subject'"'"'s sex;

classifying said measured spectrum into previously defined classes based on an estimation of stratum corneum hydration at said tissue measurement site; and

classifying said measured spectrum into previously defined classes based on skin temperature at said tissue measurement site.