Automated prostate tissue referencing for cancer detection and diagnosis

• US 9,779,283 B2
• Filed: 01/05/2012
• Issued: 10/03/2017
• Est. Priority Date: 01/05/2011
• Status: Active Grant

• Current Assignee: Board of Trustees of The University of Illinois
• Original Assignee: Board of Trustees of The University of Illinois
• Inventors: Bhargava, Rohit, Kwak, Jin Tae, Sinha, Saurabh
• Primary Examiner(s): Alam, Hosain
• Assistant Examiner(s): PHAN, TUANKHANH D
• Application Number:
  US13/344,461
• Publication Number:
  US 20120226709A1
• Time in Patent Office: 2,098 Days
• Field of Search: None
• CPC Class Codes: G06F 18/211 Selection of the most significant subset of features
• CPC Class Codes: G06F 18/251 of input or preprocessed data
• CPC Class Codes: G06T 2207/30081 Prostate
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/60 Analysis of geometric attributes
• CPC Class Codes: G06V 10/771 Feature selection, e.g. selecting representative features from a multi-dimensional feature space
• CPC Class Codes: G06V 10/803 of input or preprocessed data
• CPC Class Codes: G06V 20/693 Acquisition
• CPC Class Codes: G06V 20/695 Preprocessing, e.g. image segmentation

Abstract

This application provides to a method for identifying one or more prostate tissue samples in a database that are closest to a test prostate sample, which can be used to aid pathologists when examining prostate tissues to attain reliable and consistent diagnoses of prostate cancer. Also provided are databases and computer algorithms that can be used with such methods.

Claims

1. A method of identifying prostate tissue samples, comprising:

obtaining, by a processor, a Hematoxylin and Eosin (HE) stained image of a test prostate tissue sample;

obtaining, by the processor, an Infrared (IR) image of the test prostate tissue sample;

performing segmentation, by the processor, of lumens and nuclei in the HE stained image and the IR image of the test prostate tissue sample;

identifying, by the processor, cells, cell nuclei and lumens using the HE stained image and the IR image according to the segmentation;

selecting, by the processor, a group of morphological features from a plurality of morphological features extracted from the HE stained image and the IR image of the test prostate tissue sample identified according to the segmentation, by utilizing a minimum-redundancy-maximal-relevance (mRMR) criterion applied by the processor;

determining, by the processor, similarities between the group of morphological features extracted from the test prostate tissue sample and corresponding morphological features extracted from a plurality of prostate tissue samples, wherein HE images and IR images for the plurality of prostate tissue samples and plural morphological features extracted from the plurality of prostate tissue samples are stored in a database;

retrieving, by the processor from the database, HE images and IR images of one or more prostate tissue samples in the plurality of prostate tissue samples that are most similar to the test prostate tissue sample based on the group of morphological features selected; and

displaying the HE images and IR images of the one or more prostate tissue samples retrieved.

2. The method of claim 1, wherein the IR image is generated using a spectroscopic imaging technique.

3. The method of claim 1, wherein the plural morphological features comprise size of epithelial cells, size of a nucleus, number of nuclei, distance to lumen, distance to epithelial cell boundary, number of isolated nuclei, fraction of distant nuclei, entropy of nuclei spatial distribution, size of a lumen, number of lumens, lumen roundness, lumen distortion, lumen minimum bounding circle ratio, lumen convex hull ratio, symmetric index of lumen boundary, symmetric index of lumen area, spatial association of lumens and cytoplasm-rich regions, number of stroma cells, minimum lumen distance, minimum gland distance, ratio of lumen to epithelial cells, ratio of epithelial cells to stroma cells, ratio of cell separation, ratio of sheets of cells, degree of cell dispersion and spatial autocorrelation of cells, or any combination thereof.

4. The method of claim 1, wherein the selecting of the group of morphological features comprises:

selecting a subset of morphological features from the plurality of morphological features, wherein the determining of the group of morphological features for epithelial cells, epithelial nuclei and the lumens in the test prostate tissue sample includes analyzing the HE stained image according to the subset;

determining an additional morphological feature that is least redundant with the subset and that is correlated with a class label for the test prostate tissue sample;

adjusting the subset to include the additional morphological feature to generate an adjusted subset; and

repeating the determining of the group of morphological features for the epithelial cells, the epithelial nuclei and the lumens in the test prostate tissue sample by analyzing the HE stained image according to the adjusted subset, wherein the additional morphological feature is utilized as a starting point for the analyzing.

5. The method of claim 1, wherein the one or more prostate tissue samples that are most similar have a most similar k value to the test prostate tissue sample according to a k-nearest neighbor analysis.

6. The method of claim 1, wherein the plurality of morphological features extracted from the HE stained image and the IR image are associated with epithelial cells, epithelial nuclei and lumens.

7. The method of claim 1, wherein the database comprises information from a plurality of Gleason grade 2, 3, 4, and 5 cancer samples, benign pro static hyperplasia (BPH) samples, normal prostate samples, or combinations thereof.

8. The method of claim 1, wherein the database further comprises clinical information for the plurality of prostate tissue samples.

9. The method of claim 8, further comprising outputting the clinical information for the one or more prostate tissue samples that are most similar to a computer screen.

10. The method of claim 1, wherein selecting of the group of morphological features from the plurality of morphological features further includes utilizing a sequential floating forward selection process.

11. The method of claim 10, wherein the selecting of the group of morphological features from the plurality of morphological features includes removing one or more morphological features that improve performance.

12. The method of claim 1, performing of the segmentation for the test prostate tissue sample comprises overlaying the HE stained image and the IR image for image registration.

13. The method of claim 1, further comprising outputting immunohistochemical data from the one or more prostate tissue samples that are the most similar, wherein the immunohistochemical data is stored in the database.

14. The method of claim 1, further comprising identifying, by the processor, stroma cells according to the segmentation.

15. A system, comprising:

a microscopic imaging device;

an infrared (IR) imaging device;

a memory that stores executable instructions; and

a processor coupled with the memory, wherein the processor, responsive to executing the instructions, performs operations comprising;

capturing a digitized optical image of a Hematoxylin and Eosin (HE) stained reference sample of prostate tissue with the microscopic imaging device;

obtaining a digitized IR image of the reference sample of prostate tissue with the IR imaging device;

registering the digitized optical image and the digitized IR image, thereby forming a composite image;

segmenting the composite image into epithelium nuclei and lumen;

extracting a plurality of morphologic feature information from the segmented composite image;

storing the extracted plurality of morphologic feature information in a database;

selecting a proper subset of morphologic features from the plurality of morphologic feature information; and

determining a similarity by comparing the stored extracted plurality of morphologic feature information of the proper subset of morphologic features with corresponding morphologic feature information from a test sample of prostate tissue.

16. The system of claim 15, wherein the digitized IR image is a Fourier Transform IR image.

17. The system of claim 15, wherein the comparing is performed using a trained ranking support vector machine.

18. A non-transitory computer-readable storage medium comprising instructions which, responsive to being executed by a processor, cause the processor to perform operations comprising:

obtaining digitized optical images of prostate tissue reference samples that are Hematoxylin and Eosin (HE) stained;

obtaining corresponding digitized infrared (IR) images of the prostate tissue reference samples;

registering the digitized optical images with the corresponding digitized IR images, thereby forming composite images;

segmenting the composite images into epithelium nuclei and lumen;

extracting a plurality of morphologic feature information from the segmented composite images;

storing the extracted plurality of morphologic feature information in a database;

generating a proper subset of morphologic features from the plurality of morphologic features using a minimum-redundancy-maximal relevance criterion resulting in best performance; and

modifying the proper subset of morphologic features using a sequential floating forward selection method to add morphologic features from the plurality of morphologic features that are not in the proper subset of morphologic features that provide a greatest increase performance.

19. The non-transitory computer-readable storage medium of claim 18, wherein the operations further comprise:

extracting a plurality of morphologic feature information from a composite image of a test prostate tissue sample;

searching the database for similar morphologic feature information; and

outputting results of the search.