Classifying image features

• US 8,280,140 B2
• Filed: 04/21/2011
• Issued: 10/02/2012
• Est. Priority Date: 01/27/2005
• Status: Active Grant

• Current Assignee: Akoya Biosciences, Inc.
• Original Assignee: Cambridge Research & Instrumentation Incorporated (Revvity, Inc.)
• Inventors: Levenson, Richard, Hoyt, Clifford C., Gossage, Kirk William
• Primary Examiner(s): Tabatabai, Abolfazl
• Application Number:
  US13/091,492
• Publication Number:
  US 20110255753A1
• Time in Patent Office: 530 Days
• Field of Search: 382128-134, 382/162, 600/407, 600/476, 600/410, 356/36, 356/300, 356/369, 356/453, 356/465, 128/916, 128/920, 128/922
• US Class Current: 382/133
• 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: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06F 18/2431 Multiple classes
• CPC Class Codes: G06F 18/2433 Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 20/69 Microscopic objects, e.g. biological cells or cellular parts

Abstract

Methods are disclosed for classifying different parts of a sample into respective classes based on an image stack that includes one or more images.

Claims

1. A method assessing a tissue sample, the method comprising using one or more processors to perform the following steps:

decomposing a set of spectral images of the tissue sample into a set of unmixed images, wherein each member of the unmixed image set corresponds to a spectral contribution from a different component in the tissue sample; and

using a machine-learning classifier to classify different regions of the tissue sample into respective classes based on an image stack comprising one or more of the unmixed images.

2. The method of claim 1, wherein classifying different regions of the tissue sample comprises classifying at least some pixels in the image stack based on information that includes information derived from neighboring pixels in the image stack.

3. The method of claim 2, wherein the at least some pixels in the image stack are classified based on both spectral and spatial information derived from neighboring pixels in the image stack.

4. The method of claim 1, further comprising obtaining the set of spectral images by spectrally filtering light directed to the tissue sample or light received from the tissue sample to select light in each of a plurality of different wavelength bands, wherein each member of the set of spectral images corresponds to a different one of the wavelength bands.

5. The method of claim 4, wherein the set of spectral images comprises images that correspond to wavelength bands in red, green, and blue spectral regions.

6. The method of claim 4, wherein the obtained set of spectral images comprises bright-field images, dark-field images, fluorescence images, or luminescence images.

7. The method of claim 1, wherein classifying different regions of the tissue sample comprises:

positioning a sampling window within the image stack to select a portion of the image stack for classification, the selected portion comprising multiple pixels;

classifying the selected portion into one of several classes, wherein each of the pixels in the selected portion is provisionally classified as having the same class as that of the selected portion;

translating the sampling window to select a second portion of the image stack for classification and classifying the second portion into one of several classes, wherein each of the pixels in the second portion is provisionally classified as having the same class as that of the second portion;

repeating the translating and classifying for additional portions of the image stack until at least some of the pixels in the image stack have been provisionally classified multiple times as part of different portions selected by the sampling window; and

classifying each of at least some of the pixels that have been provisionally classified multiple times into one of the respective classes based on their multiple provisional classifications.

8. The method of claim 1, wherein the tissue sample is a tissue slice.

9. The method of claim 1, wherein the respective classes correspond different morphologies of cells within the tissue sample.

10. The method of claim 1, wherein the respective classes correspond to different biological structures within the tissue sample.

11. The method of claim 1, wherein classifying different regions of the tissue sample comprises assigning specific states or types to the different regions.

12. The method of claim 1, wherein the machine-learning classifier comprises at least one member selected from the group consisting of a neural network-based classifier, a genetic algorithm-based classifier, and a support vector machine-based classifier.

13. A system for assessing a tissue sample or a blood sample, the system, comprising:

a light source;

light conditioning optics positioned to direct light from the source to the sample;

a detector configured to detect light from the sample and record a set of spectral images;

light collecting optics positioned to direct the light from the sample to the detector; and

one or more electronic processors configured to;

decompose the set of spectral images into a set of unmixed images, wherein each member of the unmixed image set corresponds to a spectral contribution from a different component in the sample; and

use a machine-learning classifier to classify different regions of the sample into respective classes based on an image stack comprising one or more of the unmixed images.

14. The system of claim 13, wherein the light collecting optics includes one or more lenses to produce a magnified image of the sample on the detector.

15. The system of claim 13, wherein at least one of the light conditioning optics and the light collecting optics comprises a spectral filter.

16. The system of claim 15, wherein the electronic processor is further configured to obtain set of spectral images by sequentially adjusting the spectral filter to transmit light in each of several different wavelength bands, and obtaining images of the sample corresponding to each one of the wavelength bands.

17. The system of claim 16, wherein the image set comprises images that correspond to light in red, green, and blue wavelength bands.

18. The system of claim 13, wherein a central wavelength of light generated by the light source is adjustable.

19. The system of claim 13, wherein the set of spectral images are bright-field images or dark-field images.

20. The system of claim 13, wherein the set of spectral images are fluorescence images or luminescence images.

21. The system of claim 13, wherein the respective classes correspond to different morphologies of cells within the sample.

22. The system of claim 13, wherein the respective classes correspond to different biological structures within the sample.

23. The system of claim 13, wherein the machine-learning classifier comprises at least one member selected from the group consisting of a neural network-based classifier, a genetic algorithm-based classifier, and a support vector machine-based classifier.

24. The system of claim 13, wherein the electronic processor is configured to classify different regions of the sample by assigning specific states or types to the different regions.

25. The system of claim 13, further comprising a display device, and wherein the electronic processor is further configured to display classification results using the display device.

26. A method for analyzing a blood sample, the method comprising using one or more processors to perform the following steps:

decomposing a set of spectral images of the blood sample into a set of unmixed images, wherein each member of the unmixed image set corresponds to a spectral contribution from a different component in the blood sample; and

using a machine-learning classifier to classify different cells of the blood sample into respective classes based on an image stack comprising one or more of the unmixed images.

27. The method of claim 26, wherein classifying different regions of the tissue sample comprises classifying at least some pixels in the image stack based on information that includes information derived from neighboring pixels in the image stack.

28. The method of claim 27, wherein the at least some pixels in the image stack are classified based on both spectral and spatial information derived from neighboring pixels in the image stack.

29. The method of claim 26, further comprising obtaining the set of spectral images by spectrally filtering light directed to the tissue sample or received from the tissue sample to select light in each of a plurality of different wavelength bands, wherein each member of the set of spectral images corresponds to a different one of the wavelength bands.

30. The method of claim 29, wherein the set of spectral images comprises images that correspond to wavelength bands in red, green, and blue spectral regions.

31. The method of claim 29, wherein the obtained set of spectral images comprises bright-field images, dark-field images, fluorescence images, or luminescence images.

32. The method of claim 26, wherein the respective classes correspond to different cell types.

33. The method of claim 32, wherein the different cell types comprise cancer cells and non-cancer cells.

34. The method of claim 32, wherein the different cell types comprise one or more members selected from the group consisting of red blood cells, monocytes, and polymorphonuclear neutrophils.

35. The method of claim 26, wherein the respective classes correspond to cells having different morphologies.

36. The method of claim 26, wherein the machine-learning classifier comprises at least one member selected from the group consisting of a neural network-based classifier, a genetic algorithm-based classifier, and a support vector machine-based classifier.