System and method for hyper-spectral analysis
View Patent ↗An improved method and system for classifying tissue samples comprises determining a tissue type classification based on spectral data of training samples of known target classification. Denoised spectral data is generated from the spectral data based on the tissue type classification. A classifier is then trained using the denoised spectral data to classify the tissue samples.
1. An improved method for training a classifier to classify tissue samples, comprising the steps of:
determining a tissue type classification based on spectral data of training samples of known target classification;
generating denoised spectral data from said spectral data based on said tissue type classification; and
training said classifier on said denoised spectral data to provide a classification of said tissue samples.
2. The method of claim 1 , wherein the step of determining comprises the step of classifying said tissue type using a first classifier based on said spectral data.
3. The method of claim 1 , further comprising the step of classifying said tissue samples using said classifier.
4. The method of claim 1 , wherein the step of determining comprises the step of determining the tissue classification using a local disriminant basis algorithm.
5. The method of claim 1 , wherein the step of training comprises the step of training said classifier using Partial Least Squares.
6. The method of claim 1 , wherein the step of generating comprises the step of averaging nearby spectra within a fixed tissue type.
7. The method of claim 6 , further comprising the step of grouping pixels within a cellular nucleus based on said tissue classification; and wherein the step of averaging comprises the step of averaging all pixels determined to be within said cellular nucleus.
8. A method of classifying tissue samples, comprising the steps of:
determining a tissue type classification based on spectral data of training samples of known target classification;
generating denoised spectral data from said spectral data based on said tissue type classification;
training a classifier on said denoised spectral data; and
classifying said tissue samples using said classifier.
9. The method of claim 8 , wherein said training samples comprises samples containing known cancer tissue and samples not containing said known cancer tissue; and wherein the step of classifying comprises the step of classifying said tissue samples as being cancerous and non-cancerous using said classifier.
10. A system for classifying tissue samples, comprising:
a preliminary classifier for determining a tissue type classification based on spectral data of training samples of known target classification;
a processing module for generating denoised spectral data from said spectral data based on said tissue type classification; and
a training module for training a classifier on said denoised spectral data; and
wherein said classifier is operable to classify said tissue samples.
11. The system of claim 10 , wherein said preliminary classifier is operable to determine the tissue classification using a local disriminant basis algorithm.
12. The system of claim 10 , wherein said training module is operable to train said classifier using Partial Least Squares.
13. The system of claim 10 , wherein said processing module is operable to average nearby spectra within a fixed tissue type.
14. The system of claim 13 , wherein said processing module is operable to group pixels within a cellular nucleus based on said tissue classification, and average all pixels determined to be within said cellular nucleus.
15. The system of claim 10 , wherein said training samples comprises samples containing known cancer tissue and samples not containing said known cancer tissue; and wherein said classifier is operable to classify said tissue samples as being cancerous and non-cancerous.
16. A computer readable medium comprising code for classifying tissue samples, said code comprising instructions for:
determining a tissue type classification based on spectral data of training samples of known target classification;
generating denoised spectral data from said spectral data based on said tissue type classification;
training a classifier on said denoised spectral data; and
classifying said tissue samples using said classifier.
17. The computer readable medium of claim 16 , wherein said code further comprising instructions for determining the tissue classification using a local disriminant basis algorithm.
18. The computer readable medium of claim 16 , wherein said code further comprising instructions for training said classifier using Partial Least Squares.
19. The computer readable medium of claim 16 , wherein said code further comprising instructions for averaging nearby spectra within a fixed tissue type.
20. The computer readable medium of claim 19 , wherein said code further comprising instructions for grouping pixels within a cellular nucleus based on said tissue classification; and averaging all pixels determined to be within said cellular nucleus.
21. The computer readable medium of claim 16 , wherein said training samples comprises samples containing known cancer tissue and samples not containing said known cancer tissue; and wherein said code further comprising instructions for classifying said tissue samples as being cancerous and non-cancerous using said classifier.