Methods and apparatus for tailoring spectroscopic calibration models

• US 6,157,041 A
• Filed: 10/08/1999
• Issued: 12/05/2000
• Est. Priority Date: 10/13/1998
• Status: Expired due to Term

• Current Assignee: National Technology & Engineering Solutions Of Sandia LLC (Honeywell International Inc.), Rio Grande Medical Technologies Incorporated
• Original Assignee: Sandia Corporation (Martin Marietta Corporation), Rio Grande Medical Technologies Incorporated
• Inventors: Rowe, Robert K., Thomas, Edward V.
• Primary Examiner(s): Le, Que T.
• Application Number:
  US09/415,432
• Time in Patent Office: 424 Days
• Field of Search: 250/573, 250/339.09, 250/339.08, 250/339.07, 250/339.12, 250/559.1, 250/338.05, 356/39, 356/432, 356/436, 356/346, 356/303, 364/498, 364/572, 364/413.01, 364/413.02, 364/413.09, 128/633, 128/634, 128/664, 128/665
• US Class Current: 250/573
• 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: G01J 3/28 Investigating the spectrum using colour filters G01J3/51
• CPC Class Codes: G01N 21/359 using near infrared light

Abstract

A method and apparatus for non-invasively measuring a biological attribute, such as the concentration of an analyte, particularly a blood analyte in tissue such as glucose. The method utilizes spectrographic techniques in conjunction with an improved subject-tailored calibration model. In a calibration phase, calibration model data is modified to reduce or eliminate subject-specific attributes, resulting in a calibration data set modeling within--subject physiological variation, sample location, insertion variations, and instrument variation. In a prediction phase, the prediction process is tailored for each target subject separately using a minimal number of spectral measurements from each subject.

Claims

1. A method for generating a prediction result for use on a specific subject to predict a biological attribute of that subject using spectroscopy as a surrogate indirect measurement for a direct measurement of said biological attribute, said method comprising the steps of:

(a) acquiring a calibration data set and modifying said calibration data set in a manner that reduces the spectral variation due to subject specific attributes;

(b) generating a model by applying multivariate analysis to said. modified calibration data set; and

(c) using a prediction process to predict an unknown amount of said biological attribute in a target spectroscopic measurement from said specific subject, said prediction process utilizing said model in conjunction with one or more reference measurements.

2. The method of claim 1, wherein said reference measurements are spectroscopic measurements.

3. The method of claim 1, wherein said reference measurements include both spectroscopic measurements and direct measurements from said specific subject.

4. The method of claim 3, wherein said direct measurements are a blood analyte measurement.

5. The method of claim 1, wherein said calibration data is obtained from a series of spectroscopic measurements for a number of subjects with corresponding direct measurements of said biological attribute.

6. The method of claim 5, wherein modifying said calibration data set to reduce subject specific spectral attributes in said calibration data set includes forming the mean indirect measurement and mean direct measurement for each subject from which calibration data is obtained based on the number of measurements from that subject followed by mean centering the indirect measurement by subject by subtracting the mean indirect measurement from each subject from each indirect measurement, and meancentering the direct measurement by subtracting the mean direct measurement from each direct measurement for each subject.

7. The method of claim 5, wherein modifying said calibration data set to reduce subject specific spectral attributes for each subject in said calibration data set includes subtracting the mean of the first S indirect measurements and direct measurements from a particular subject from each of the subject'"'"'s indirect measurements, where S is less than the total number of indirect and direct measurements made on that subject in generating the calibration data set.

8. The method of claim 5, wherein modifying said calibration data set to reduce subject specific spectral attributes for each subject in said calibration data set includes subtracting the mean of the S nearest in time indirect and direct measurements of a subject from each of the subject'"'"'s indirect and direct measurements, where S is less than the total number of indirect measurements used in forming the calibration data set.

9. The method of claim 5, wherein modifying said calibration data set to reduce subject specific spectral attributes for each subject in said calibration data set includes subtracting each of a subjects direct and indirect measurements from every other direct and indirect measurement made on that subject in a round-robin fashion.

10. The method of claim 5, further including providing a stored spectral library wherein modifying said calibration data set to reduce subject specific spectral attributes for each subject in said calibration data set includes subtracting a combination of spectral data from a stored spectral library based on matching said series of spectroscopic measurements from each of said number of subjects with a stored measurement in said spectral library.

11. The method of claim 5, wherein modifying said calibration data set to reduce subject specific spectral attributes for each subject in said calibration data set includes subtracting simulated data from indirect measurements, said simulated data derived from prior modeling of spectral attributes.

12. The method of claim 3, wherein said prediction process utilizes said model and said references measurements to calculate a prediction of the direct measurement and utilizes the difference between the prediction of the direct measurement and the direct measurement of said biological attribute to estimate a correction factor.

13. The method of claim 1, wherein said reference measurements are replaced by matched measurements.

14. The method of claim 13, wherein said matched measurements are selected from said spectral library by calculating a measure of the difference between said target spectroscopic measurement and said library spectra.

15. A method of generating a calibration model that is essentially free from subject specific effects comprising building a generic model by:

(a) obtaining a series of indirect measurements from a number of subjects, and obtaining a direct measurement for each subject corresponding to each indirect measurement;

(b) forming the mean indirect measurement and the mean direct measurement for each subject based on the number of measurements from said each subject;

(c) meancentering the series of indirect measurements by subject by subtracting the mean indirect measurement from each subject from each indirect;

measurement, and meancentering the direct measurement by subtracting the mean direct measurement from each direct measurement for each subject; and

(d) forming a generic calibration model from the meancentered direct and indirect measurements.

16. The method if claim 15, further comprising tailoring the generic calibration model to a specific subject.

17. The method of claim 15, wherein said indirect measurements are spectral measurements.

18. The method of claim 15, wherein the measurements are made on a single measurement device, whereby the generic calibration model is to be utilized in said single measurement device.

19. The method of claim 15, wherein said direct measurements are of a desired blood component and said desired blood component is measured by invasively removing blood from said each subject and analyzing the blood for said desired blood component.

20. The method of claim 19, wherein the desired component is glucose.

21. The method of claim 16, wherein the tailoring step further comprises:

(a) making a direct measurement, G_ref, and at least one indirect measurement Y_ref, of the specific subject;

(b) using the generic calibration model with Y_ref to obtain a raw prediction, P_o, of the physical characteristic.

22. The method of claim 21, wherein the measurements are made on a single measurement device, whereby the generic calibration model is to be utilized in said single measurement device.

23. The method of claim 21, further comprising:

making a plurality of indirect measurements of the specific subject, Y_new ;

using the generic calibration model with Y_new to obtain an untailored prediction, P_new ; and

predicting the physical characteristic G_new for the subject as a function of P_new, P_o, and G_ref.

24. The method of claim 23, wherein G_new is also a function of a known scale factor.

25. The method of claim 21, wherein the tailoring step further comprises:

determing P₀ and G_ref according to the method of claim 22 once with the specific subject at a relatively high level of the physical characteristic and once with the specific subject at a relatively low level of the physical characteristic; and

determining a scale factor based on P₀ and G_ref at high and low levels.

26. A method for generating a prediction result for use on a specific subject to predict a biological attribute of said specific subject using spectroscopy as a surrogate indirect measurement for a direct measurement of said biological attribute, said method comprising the steps of:

(a) using a modified calibration data set previously processed in a manner that reduces the spectral variation due to subject specific attributes;

(b) generating a calibration model through application of a multivariate algorithm that uses a composite calibration data set formed by combining the modified calibration data with two or more reference measurements; and

(c) predicting an unknown amount of said biological attribute in a target spectroscopic measurement utilizing said calibration model.

27. The method of claim 26, wherein said composite calibration data is created by combining in a linear manner reference measurements with said calibration data, the combining process to include both reference spectra and reference analyte measurements.

28. The method of claim 26, wherein said calibration data is obtained from a series of spectroscopic measurements for a number of subjects with corresponding direct measurements of said biological attribute.

29. The method of claim 28, wherein modifying said calibration data set to reduce subject specific spectral attributes for each subject in said calibration data set includes forming the mean indirect measurement and mean direct measurement for each subject based on the number of measurements from that subject followed by mean the mean centering the indirect measurement by subject by subtracting the mean indirect measurement from each subject from each indirect measurement, and meancentering the direct measurement by subtracting the mean direct measurement from each direct measurements for each subject.

30. The method of claim 28, wherein said calibration data with reduced subject specific spectral attributes for each subject is modified by subtracting the mean of the first S indirect measurements and direct measurements from a particular cane of said number of subjects from each of the subject'"'"'s indirect measurements, where S is less than the total number of indirect and direct measurements made on said particular one of said number of subjects in generating the calibration data set.

31. The method of claim 28, wherein said calibration data with reduced subject specific spectral attributes for each subject is modified by subtracting the mean of the S nearest in time indirect and direct measurements of a subject from each of the subject'"'"'s indirect and direct measurements, where S is less than the total number of indirect measurements used in forming the calibration data set.

32. The method of claim 28, wherein said calibration data with reduced subject specific spectral attributes for each subject is modified by subtracting each of a subjects direct and indirect measurements from every other direct and indirect measurement made on that subject on a round-robin fashion.

33. The method of claim 28, further including providing a stored spectral library wherein said calibration data with reduced subject specific spectral attributes for each subject is modified by subtracting a combination of spectral data from a stored spectral library based on matching said series of spectroscopic measurements from each of said number of subjects with a stored measurement in said spectral library.

34. The method of claim 28, wherein modifying said calibration data set to reducee subject specific spectral attributes for each subject in said calibration data set includes subtracting simulated data from indirect measurements, said simulated data derived from prior modeling of spectral attributes.

35. The method of claim 26, wherein said reference measurements are replaced by matched measurements.

36. The method of claim 35, wherein said matched measurements are selected from said spectral library by calculating a measure of the difference between said target spectroscopic measurement and said library spectra.

37. A method for predicting a measure of a biological attribute for a specific subject, comprising:

(a) obtaining a calibration data set of direct and indirect measurements of the biological attribute from a plurality of calibration subjects, wherein the calibration data set has been modified to reduce variations therein due to subject specific attributes for each calibration subject;

(b) developing a subject-specific calibration model from said modified calibration data set tailored for the specific subject with at least one reference measurement of the biological attribute from the specific subject;

(c) obtaining at least one indirect measurement of the biological attribute for the specific subject; and

(d) using the said subject-specific calibration model and said "at least one" indirect measurement of the biological attribute for the specific subject to predict a measure of the biological attribute in the specific subject.

38. The method of claim 37, wherein modifying said calibration data set to reduce subject specific spectral attributes for each calibration subject in said calibration data set includes subtracting simulated data from indirect measurements, said simulated data derived from prior modeling of spectral attributes.

39. The method of claim 37, further including providing a stored spectral library wherein said calibration data with reduced subject specific spectral attributes for each calibration subject is modified by subtracting a combination of spectral data from a stored spectral library based on matching the calibration subject'"'"'s indirect measurement with a stored measurement in said spectral library.

40. The method of claim 37, wherein the calibration data set is modified to reduce variations in the direct measurements of the biological attribute due to subject specific attributes for each subject.

41. The method of claim 37, wherein the calibration data set is modified to reduce variations in the indirect measurements of the biological attributes due to subject specific attributes for each calibration subject.

42. The method of claim 37, further including forming a prediction model from the direct and indirect measurements of the biological attributes for each subject.

43. The method of claim 37, wherein the specific subject is not one of the calibration subjects.

44. The method of claim 37, wherein modifying said calibration data set to reduce subject specific spectral attributes for each calibration subject in said calibration data set includes forming the mean indirect measurement and mean direct measurement for each calibration subject based on the number of measurements from that calibration subject followed by mean centering the indirect measurement by subject by subtracting the mean direct measurement from each direct measurement for each calibration subject.

45. The method of claim 37, wherein said calibration data with reduced subject specific spectral attributes for each subject is modified by subtracting the mean of the first S indirect measurements and direct measurements from a particular subject from each of the subject'"'"'s indirect measurements, where S is less than the total number of indirect and direct measurements made on that subject in generating the calibration data set.

46. The method of claim 37, wherein said calibration data with reduced subject specific spectral attributes for each subject is modified by subtracting the mean of the S nearest in time indirect and direct measurements of a subject from each of the subject'"'"'s indirect and direct measurements, where S is less than the total number of indirect measurements used in forming the calibration data set.

47. The method of claim 37, wherein said calibration data with reduced subject specific spectral attributes fore each subject is modified by subtracting each of a subjects direct and indirect measurements from every other direct and indirect measurement made on that subject on a round-robin fashion.

48. A non-invasive method for measuring a biological attribute in human tissue of a specific subject comprising the steps of:

(a) providing an apparatus for measuring infrared absorption, said apparatus including an energy source emitting infrared energy at multiple wavelengths operatively connected to an input element, said apparatus further including an output element operatively connected to a spectrum analyzer;

(b) coupling said input and output elements to said human tissue;

(c) irradiating said tissue through said input element with multiple wavelengths of infrared energy with resulting differential absorption of at least some of said wavelengths;

(d) collecting at least a portion of the non-absorbed infrared energy with said output element followed by determining the intensities of said wavelengths of the non-absorbed infrared energy; and

(e) predicting the biological attribute of said specific subject utilizing a model, wherein said subject specific prediction method uses spectroscopic variation from multiple subjects and one or more reference measurements from said specific subject, each of said reference measurements including spectroscopic and corresponding direct measurement of said biological attribute.

49. A quantitative analysis instrument for non-invasive measurement of a biological attribute in human tissue of a specific subject, said instrument comprising:

(a) a source of multiple wavelengths of infrared energy;

(b) an input sensor element for directing said wavelengths of infrared energy into said tissue and an output sensor element for collecting at least a portion of the non-absorbed diffusely reflected infrared energy from said tissue, said input and said output sensors adapted to couple to the surface of said tissue;

(c) at least one detector for measuring the intensities of at least a portion of said wavelengths collected by said output sensor element; and

(d) electronics for processing said measured intensities and indicating a value for said biological attribute, said electronics including a processing method incorporated therein, said method utilizing calibration data which has been developed in a manner that reduces subject specific spectral attributes and said method utilizes one or more reference measurements from said specific subject.

50. The instrument of claim 49, wherein the electronics for processing one or more reference measurements from said specific subjects uses said reference measurements to remove the specific subject attributes from said measured intensities.

51. The instrument of claim 49, wherein the processing method combines said reference measurements with said calibration data to create a subject specific model.

52. The instrument of claim 49, wherein multiple subjects are used for development of the calibration data.

53. The instrument of claim 52, wherein the multiple subject spectra are processed to reduce subject specific attributes.

54. The instrument of claim 53, within the multiple subject spectra with reduced subject specific attributes are created by subtracting some linear combination of each subject'"'"'s spectra from the same subject'"'"'s spectra.

55. An instrument for the non-invasive measurement of a biological attribute for a specific subject, comprising:

(a) a memory adapted to store a calibration data set of direct and indirect measurements of the biological attribute obtained from a plurality of calibration subjects, wherein the calibration data set has been modified to reduce variations therein due to subject specific attributes for each calibration subject;

(b) means for developing a subject-specific calibration model from said modified calibration data set wherein said data set is tailored for the specific subject with at least one reference measurement of the biological attribute from the specific subject;

(c) means for obtaining at least one indirect measurement of the biological attribute from the specific subject; and

(d) means for obtaining a measurement of the biological attribute for the specific subject using the subject-specific calibration model and at least one indirect measurement of the biological attribute for the specific subject.

56. The instrument of claim 55, wherein the electronics for processing one or more references measurement from said specific subjects uses a process that incorporates both the reference measurements and said calibration data to generate prediction results from said measured intensities.

57. The instrument of claim 55, wherein the electronics for processing one or more references measurement from said specific subjects uses a process that incorporates both the reference measurements and said measured intensities for generation of a prediction result.

58. The instrument of claim 55, wherein the reference measurement is a direct measurement of the biological attribute for the specific subject.

59. The instrument of claim 55, wherein the calibration data comprises spectroscopic measures of the biological attributes.

60. The instrument of claim 55, wherein the calibration data set reduced variations in subject specific attributes are created by subtracting a linear combination of each subject'"'"'s spectra from the same subject'"'"'s spectra.

61. A method for predicting a variable, comprising:

(a) obtaining a calibration data set of direct measurements and indirect spectral measurements of the variable from a plurality of environments, wherein the calibration data set has been modified to reduce variations therein due to environment-specific attributes for each environment;

(b) developing a environment-specific calibration model from said modified calibration data set tailored for the specific environment with at least one reference measurement of the variable from the specific environment;

(c) obtaining at least one indirect measurement of the variable for the specific environment; and

(d) using the said environment-specific calibration model and said "at least one" indirect measurement of the variable for the specific environment to predict a measure of the variable in the specific environment.

62. The variable of claim 61, wherein the variable is a chemical or biological pollutant.

63. The environment of claim 61, wherein the environment is a chemical reactor.

64. The environment of claim 61, wherein the environment is a geophysical environment.