Systematic wavelength selection for improved multivariate spectral analysis

• US 5,435,309 A
• Filed: 08/10/1993
• Issued: 07/25/1995
• Est. Priority Date: 08/10/1993
• Status: Expired due to Term

• Current Assignee: Sandia Corporation (Martin Marietta Corporation)
• Original Assignee: Sandia Corporation (Martin Marietta Corporation)
• Inventors: Haaland, David M., Thomas, Edward V., Robinson, Mark R.
• Primary Examiner(s): Sykes, Angela D.
• Application Number:
  US08/104,857
• Time in Patent Office: 714 Days
• Field of Search: 128/633-635, 128/632, 128/664-667, 356/39-41, 356/51, 356/300, 356/302-303, 356/306, 356/320, 356/432, 356/436-437
• US Class Current: 600/310
• CPC Class Codes: A61B 5/14532 for measuring glucose, e.g. by tissue impedance measurement
• CPC Class Codes: A61B 5/14546 for measuring analytes not otherwise provided for, e.g. ions, cytochromes
• CPC Class Codes: A61B 5/1455 using optical sensors, e.g. spectral photometrical oximeters
• CPC Class Codes: A61B 5/14552 Details of sensors specially adapted therefor
• CPC Class Codes: A61B 5/1464 specially adapted for foetal tissue
• CPC Class Codes: A61B 5/6826 Finger
• CPC Class Codes: A61B 5/6838 Clamps or clips
• CPC Class Codes: A61B 5/6843 Monitoring or controlling sensor contact pressure
• CPC Class Codes: G01N 21/31 Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry G01N21/72 takes precedence

Abstract

Methods and apparatus for determining in a biological material one or more unknown values of at least one known characteristic (e.g. the concentration of an analyte such as glucose in blood or the concentration of one or more blood gas parameters) with a model based on a set of samples with known values of the known characteristics and a multivariate algorithm using several wavelength subsets. The method includes selecting multiple wavelength subsets, from the electromagnetic spectral region appropriate for determining the known characteristic, for use by an algorithm wherein the selection of wavelength subsets improves the model'"'"'s fitness of the determination for the unknown values of the known characteristic. The selection process utilizes multivariate search methods that select both predictive and synergistic wavelengths within the range of wavelengths utilized. The fitness of the wavelength subsets is determined by the fitness function F=ƒ(cost, performance). The method includes the steps of: (1) using one or more applications of a genetic algorithm to produce one or more count spectra, with multiple count spectra then combined to produce a combined count spectrum; (2) smoothing the count spectrum; (3) selecting a threshold count from a count spectrum to select these wavelength subsets which optimize the fitness function; and (4) eliminating a portion of the selected wavelength subsets. The determination of the unknown values can be made: (1) noninvasively and in vivo; (2) invasively and in vivo; or (3) in vitro.

Claims

1. In a method for use with optical instrumentation for determining one or more unknown values of at least one known characteristic by an optical measurement, said method including the steps of:

(a) irradiating said material having said unknown values of said known characteristic with electromagnetic energy including at least several wavelengths so that there is differential absorption of at least some of said wavelengths by said material as a function of said wavelengths and said characteristic, said differential absorption causing intensity variations of said wavelengths incident from said material as a function of said wavelengths and said unknown values of said known characteristic;

(b) measuring said intensity variations from said material; and

(c) calculating said unknown values of said known characteristic in said material from said measured intensity variations utilizing an algorithm and a model, said algorithm being capable of using all independent sources of intensity variations v. wavelengths information obtained from irradiating a set of samples with a range of wavelengths in which said values of said known characteristic are known, said algorithm also being capable of using more wavelengths than samples in said set of samples, said model constructed from said set of samples and being a function of said known values of said characteristic and said intensity variations v. wavelengths information obtained from irradiating said set of samples, the improvement comprising selecting multiple variable subsets for generation and use in an improved model, each of said subsets containing one or more variables, said model being improved by selecting said multiple variable subsets from the set of instrument variables and wherein said algorithm with said improved model improves the fitness for said determination of said unknown values of said known characteristic, said selection process utilizing multivariate search methods that select both predictive and synergistic variables.

2. The method as set forth in claim 1, wherein said variables include said wavelengths used by said optical instrumentation to irradiate said material.

3. The method as set forth in claim 2, wherein said wavelength subset selection step is made independent of the knowledge of the spectral features of the characteristics of said material.

4. The method as set forth in claim 1, wherein said fitness (F) is defined as:

space="preserve" listing-type="equation">F=ƒ

(cost, performance).

5. The method as set forth in claim 4, wherein factors utilized in determining said cost contribution to said fitness are selected from the group including:

measurement time, instrument resolution, wavelength range, and number of wavelength subsets measured.

6. The method as set forth in claim 4, wherein factors utilized in determining said performance contribution to said fitness are selected from the group including:

SEP, outlier detection, range of said values of said known characteristic covered by said model, the robustness of said model, and the ease of transferability of said model.

7. The method as set forth in claim 4, wherein said variable subset selection process is made utilizing a genetic algorithm.

8. The method as set forth in claim 7, further including the step of generating a count spectrum.

9. The method as set forth in claim 8, wherein said variable subset selection process further includes the step of selecting a threshold count from said count spectrum to select said variable subsets.

10. The method as set forth in claim 9, wherein said variable subset selection process includes the step of eliminating a portion of said selected variable subsets.

11. The method as set forth in claim 8, wherein said variable subset selection process further includes the step of smoothing said count spectrum.

12. The method as set forth in claim 11, wherein said variable subset selection process further includes the step of selecting a threshold count from said count spectrum to select said variable subsets.

13. The method as set forth in claim 12, wherein said variable subset selection process includes the step of eliminating a portion of selected variable subsets.

14. The method as set forth in claim 7, wherein said variable subset selection process is performed by using multiple applications of said genetic algorithm.

15. The method as set forth in claim 14, further including the step of generating multiple count spectra, and wherein said count spectra are combined to produce a combined count spectrum.

16. The method as set forth in claim 15, wherein said variable subset selection process further includes the step of selecting a threshold count from said combined count spectrum to select said variable subsets.

17. The method as set forth in claim 16, wherein said variable subset selection process includes the step of eliminating a portion of said selected variable subsets.

18. The method as set forth in claim 15, wherein said variable subset selection process further includes the step of smoothing said combined count spectrum.

19. The method as set forth in claim 18, wherein said variable subset selection process further includes the step of selecting a threshold count from said combined count spectrum to select said variable subsets.

20. The method as set forth in claim 19, wherein said variable subset selection process includes the step of eliminating a portion of said selected variable subsets.

21. The method as set forth in claim 1, wherein said algorithm is selected from the group including PLS, PLS2, PCR, CLS, Q-matrix, cross-correlation, Kalman filtering, neural networks, and continuum regression.

22. The method as set forth in claim 1, wherein said determination is made for a known characteristic of a solid material.

23. The method as set forth in claim 1, wherein said determination is made for a known characteristic of a liquid.

24. The method as set forth in claim 1, wherein said determination is made for a known characteristic of a gas.

25. The method as set forth in claim 1, wherein said variables utilized in said variable selection are selected from the group including:

physical properties, chemical properties, temperature, the number of factors used in the model, and wavelengths.