Method and system for using a weighted response

• US 20030186461A1
• Filed: 03/29/2002
• Published: 10/02/2003
• Est. Priority Date: 03/29/2002
• Status: Active Grant

• Current Assignee: Smiths Detection Incorporated
• Original Assignee: Cyrano Bioscienes Incorporated
• Inventors: Zhang, Shou-Hua, Vicic, Michael A., Hsiung, Chang-Meng, Boehr, Christopher K.
• Granted Patent: US 7,129,095 B2
• Time in Patent Office: Days
• US Class Current: 436/181
• CPC Class Codes: G01N 33/0034 comprising neural networks or related mathematical techniques
• CPC Class Codes: Y10T 436/25875 Gaseous sample or with change of physical state

Abstract

The present invention is related to the sensing of analytes and in particular, to methods of determining, resolving, identifying, or quantitating such analytes once detected. This invention provides systems and methods for identifying analytes, comprising using an importance index to give greater weight to the responses from sensors that measure characteristics most useful for identification of analytes. In other aspects, the systems and methods are useful for increasing the stability of electronic nose systems by prolonging the predictive capability of the training set of known analytes. In still other aspects, the systems and methods are useful for detecting and responding to events correlated with the presence of an analyte.

Claims

1. A method for identifying an analyte, said method comprising:

a) disposing a plurality of d sensors in an array, each sensor having a different response to said analyte;

b) exposing said d sensors to said analyte to generate a response vector for said analyte;

c) applying an importance index vector to said response vector to generate a weighted response vector; and

d) comparing said weighted response vector to a weighted training set and thereafter selecting a match to identify said analyte.

2. The method of claim 1, wherein said importance index vector is generated after a training mode conducted on said d sensors and before identification of said analyte.

3. The method of claim 2, wherein said training mode comprises:

a) exposing said d sensors two or more times to a plurality of known analytes and measuring the responses from each sensor to each exposure to create a training set, wherein said training set contains a response vector for each exposure to each of said plurality of known analytes;

b) normalizing and scaling said response vector for each exposure to each of said plurality of known analytes; and

c) performing cross-validation on said training set.

4. The method of claim 2, wherein said importance index vector is the collection of importance indices for each sensor and the weighted training set is generated by application of the importance index to each said response vector for each exposure to each of said plurality of known analytes in a training set.

5. The method of claim 4, wherein said importance index for a particular sensor is calculated from all the responses for that sensor in a training set as follows:

6. The method of claim 5, wherein s² _{OverallAmongAnalytes} equals:

7. The method of claim 1, wherein said importance index vector is generated after a training mode conducted on a device different from said d sensors.

8. The method of claim 4, wherein the application of said importance index vector to the response vector to generate a weighted response vector comprises:

a) raising the importance index for each sensor to a certain power;

b) multiplying each response in the response vector by the importance index for the corresponding sensor raised to a certain power to yield a weighted response; and

c) assembling the weighted responses to generate a weighted response vector.

9. The method of claim 4, wherein the application of said importance index vector to the training set to generate a weighted training set comprises:

a) raising the importance index for each sensor to a certain power;

b) multiplying each response in the response vector for the training set by said importance index for the corresponding sensor raised to a certain power to yield a weighted response; and

c) assembling the weighted responses to generate a weighted response vector.

10. The method of claim 8, wherein said importance index is raised to the first power.

11. The method of claim 1, further comprising noise filtering, noise reduction, normalization, mean-centering, variance scaling, autoscaling, or combinations thereof.

12. The method of claim 11, wherein said noise filtering, noise reduction, normalization, mean-centering, variance scaling, or autoscaling occur prior to calculation of the importance index.

13. The method of claim 1, wherein said weighted response vector is compared to the weighted response vectors in the training set using a pattern recognition algorithm selected from the group consisting of Principal Component Analysis, Fisher Linear Discriminant Analysis, Partial Least Squares Discriminant Analysis, Support Vector Machine Analysis, Kernel Fisher Discriminant Analysis, Soft Independent Modeling of Class Analogy, K-Nearest Neighbors, Canonical Discriminant Analysis, neural networks, genetic algorithms, and fuzzy logic.

14. The method of claim 13, wherein said response vectors are compared using Canonical Discriminant Analysis.

15. The method of claim 14, wherein said response vectors are compared using the K Nearest Neighbors algorithm.

16. The method of claim 1, wherein said analyte is an odor.

17. The method of claim 1, wherein each of said sensors has an electrical response to said analyte.

18. The method of claim 1, wherein said d sensors are disposed on a single substrate.

19. The method of claim 1, wherein said sensor array comprises at least one member selected from the group consisting of thermal sensors, radiation sensors, mechanical sensors, magnetic sensors, and chemical sensors.

20. The method of claim 1, wherein said sensor array comprises at least one member selected from the group consisting of inorganic metal oxide semiconductor sensors, mass-sensitive piezoelectric sensors, nonconducting/conducting regions sensors, surface acoustic wave devices, tin oxide detectors, conducting organic polymers, dye-impregnated polymer films on fiber optic detectors, polymer-coated micromirrors, quartz crystal microbalances, electrochemical gas detectors, chemically sensitive field-effect transistors, and carbon black-polymer composite chemiresistors.

21. A system for identifying an analyte, said system comprising:

a) a sensor array having a plurality of d sensors wherein each sensor has a different response to said analyte; and

b) a sensor array interface for detecting the response of each sensor upon exposure of said d sensors to said analyte; and

c) a computer for generating a response vector for each said exposure, applying an importance index vector to said response vector to generate a weighted response vector, and comparing said weighted response vector to the weighted training set, and thereafter selecting the best match to identify said analyte.

22. A method for identifying an unknown analyte, comprising:

a) disposing a plurality of d sensors in a array, each having a different response to different analytes;

b) exposing said d sensors two or more times to a plurality of known analytes and measuring the responses from each sensor to each exposure to create a first training set, wherein said first training set contains a response vector for each exposure to each of said plurality of known analytes;

c) generating a first importance index for each sensor;

d) selecting a subset of sensors having first importance indices above a certain threshold;

e) generating a second training set, wherein said second training set consists of all the response vectors in the first training set but with responses to those sensors not passing the selection criterion removed;

f) calculating a second importance index based on the second training set and assembling a second importance index vector;

g) applying the second importance index vector to the second training set to generate a second weighted training set;

h) exposing said subset of sensors to said unknown analyte to generate a response vector for said unknown analyte;

i) applying said second importance index vector to said response vector for the unknown analyte to obtain a weighted response vector; and

j) comparing said weighted response vector to the the second weighted training set, and then selecting the best match thereby identifying said analyte.

23. A method for identifying an unknown analyte, said method comprising:

a) disposing a plurality of d sensors in a array, each having a different response to different analytes;

b) exposing said d sensors two or more times to a plurality of known analytes and measuring the responses from each sensor to each exposure to create a first training set, wherein said first training set contains a response vector for each exposure to each of said plurality of known analytes;

c) generating an first importance index for each sensor;

d) identifying a subset of sensors with importance indices above a certain threshold;

e) generating a second training set, wherein said second training set consists of all the response vectors in the first training set but with responses to those sensors not passing the selection criterion removed;

f) exposing said subset of sensors to said unknown analyte to generate a response vector; and

g) comparing said response vector for the unknown analyte to response vectors in the second training set, and thereafter selecting the best match to identify said unknown analyte.

24. A method for quantifying the concentration of an analyte, said method comprising:

a) disposing a plurality of d sensors in an array, each sensor having a different response to changes in concentration of an analyte;

b) exposing said d sensors to said analyte to generate a response vector for said analyte;

c) applying an importance index vector to said response vector to generate a weighted response vector; and

d) comparing said weighted response vector to a weighted training set, wherein said training set comprises the responses to said analyte at different concentrations, and thereafter selecting a match to quantify the concentration of said analyte.

25. A method for detecting and responding to an event, said method comprising:

a) disposing a plurality of d sensors in an array, each sensor having a different response to an analyte, wherein the presence of said analyte is correlated with the occurrence of an event;

b) exposing said d sensors to said analyte to generate a response vector for said analyte;

c) applying an importance index vector to said response vector to generate a weighted response vector;

d) comparing said weighted response vector to a weighted training set;

e) selecting a match to identify said analyte; and

f) responding to the event correlated with the presence of said analyte.

26. A method for increasing the stability of an electronic nose system, said method comprising:

a) disposing a plurality of d sensors in an array, each sensor having a different response to different analytes;

b) exposing said d sensors two or more times to a plurality of known analytes and measuring the responses from each sensor to each exposure to create a training set, wherein said training set contains a response vector to each of said plurality of known analytes;

c) calculating an importance index for each sensor and generating an importance index vector; and

d) applying said importance index vector to said training set to create a weighted training set, thereby increasing the stability of an electronic nose system.

27. A method for prolonging the predictive capability of a training set for a sensor array, said method comprising:

a) disposing a plurality of d sensors in an array, each sensor having a different response to different analytes;

b) exposing said d sensors two or more times to a plurality of known analytes and measuring the responses from each sensor to each exposure to create a training set, wherein said training set contains a response vector for each exposure to each of said plurality of known analytes;

c) calculating an importance index for each sensor and generating an importance index vector; and

d) applying said importance index vector to said training set to create a weighted training set, thereby prolonging the predictive capability of said training set.