Profiling of Pathology Images for Clinical Applications

• US 20180374210A1
• Filed: 11/15/2016
• Published: 12/27/2018
• Est. Priority Date: 11/17/2015
• Status: Active Grant

• Current Assignee: Board of Trustees of the Leland Stanford Junior University (Stanford University)
• Original Assignee: Board of Trustees of the Leland Stanford Junior University (Stanford University)
• Inventors: Barker, Jocelyn E., Rubin, Daniel L.
• Granted Patent: US 10,943,345 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/10056 Microscopic image
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/20021 Dividing image into blocks, subimages or windows
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/30016 Brain
• CPC Class Codes: G06T 2207/30024 Cell structures in vitro; Tissue sections in vitro
• CPC Class Codes: G06T 2207/30028 Colon; Small intestine
• CPC Class Codes: G06T 2207/30061 Lung
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/13 Edge detection
• CPC Class Codes: G06T 7/90 Determination of colour characteristics
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: G06V 20/695 Preprocessing, e.g. image segmentation
• CPC Class Codes: G06V 20/698 Matching; Classification
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images

Abstract

Provided are automated (computerized) methods and systems for analyzing digitized pathology images in a variety of tissues potentially containing diseased or neoplastic cells. The method utilizes a coarse-to-fine analysis, in which an entire image is tiled and shape, color, and texture features are extracted in each tile, as primary features. A representative subset of tiles is determined within a cluster of similar tiles. A statistical analysis (e.g. principal component analysis) reduces the substantial number of “coarse” features, decreasing computational complexity of the classification algorithm. Afterwards, a fine stage provides a detailed analysis of a single representative tile from each group. A second statistical step uses a regression algorithm (e.g. elastic net classifier) to produce a diagnostic decision value for each representative tile. A weighted voting scheme aggregates the decision values from these tiles to obtain a diagnosis at the whole slide level.

Claims

1. A computerized method for classifying tissue characteristics in digital pathology images, comprising the steps of:

(a) obtaining a digital pathology image of a tissue from a subject;

(b) dividing the digital pathology image into tiles;

(c) extracting primary features from the tiles in step (b), the primary features comprising shape, color, and texture features in the image;

(d) grouping similar tiles into a number of sets based on similarity of the primary features;

(e) selecting a representative tile from each set in step (d);

(f) extracting secondary features from selected tiles from step (e), wherein the secondary features refine primary features;

(g) assigning values to selected tiles, based on secondary features; and

(h) comparing the values in step (g) to values in a reference.

2. The method according to claim 1, comprising generating a classification of tissue in the image based on the comparing the values in step (g) to values in a reference.

3. The method according to claim 2, wherein the classification indicates a disease condition or prognosis.

4. The method according to claim 1, wherein the tissue is selected from the group consisting of:

brain, lung, breast, thyroid, blood, prostate tissue, duodenum, and colon.

5. The method according to claim 4, wherein the tissue is brain tissue.

6. The method according to claim 1, wherein the slides are digitally scanned at multiple resolutions between from about 2.5-40×

.

7. The method according to claim 1, wherein the step of (a) obtaining a digital pathology image comprises obtaining an image of a tissue stained with two different stains, one basic and one acidic, whereby DNA structures and RNA structures are stained by the basic stain and proteins and cytoplasmic structures are stained by the acidic stain.

8. The method according to claim 1, wherein the step of (b) dividing substantially the entire pathology image into tiles comprises dividing into between 5 and 20 tiles, and each tile has a size of about 1024 by 1024 pixels (1,048,576 pixels), or a pixel count between about 500,000 pixels to 2 million pixels.

9. The method according to claim 1, further comprising a step of deconvoluting colors from the two different stains, prior to extracting primary features (c).

10. The method according to claim 1, wherein the step of (c) extracting primary features comprises determining at least one of nuclear segmentation, an unaligned Reisz feature, and a Haralick texture feature.

11. The method according to claim 1, wherein the step of (d) grouping similar tiles comprises conducting one or both of a principal component analysis (PCA) to reduce the number of primary features and a K-means clustering.

12. The method according to claim 1, wherein the step of (e) selecting a representative tiles comprises determining a single representative tile whose values are the closest to the centroid of the cluster.

13. The method according to claim 1, wherein step of (f) extracting secondary features comprises (i) determining primary features in greater detail as in primary features and at least one of (ii) aligning Reisz features and (iii) determining more skeleton features of the nuclear morphology.

14. The method according to claim 1, wherein the step of (g) assigning a value to each selected tile comprises:

generating a classification using a regression model, weighted voting, or both.

17. The method according to claim 2, wherein generating a classification comprises a classification of a type of cancer.

18. The method according to claim 17, wherein the classification distinguishes a type of brain cancer.

19. The method according to claim 18, wherein the classification is one of lower grade glioma and glioblastoma multiforme.

20. The method according to claim 2, wherein generating a classification comprises a classification of gender, and the tissue is brain tissue from a subject having a brain cancer.

23. A computer system for classifying tissue characteristics in digital pathology images, comprising a computer-readable storage device containing a set of instructions that causes the computer system to perform the steps of the method according to claim 1.

24. A computer-readable storage medium containing instructions for carrying out the method according to claim 1.

15-16. -16. (canceled)

21-22. -22. (canceled)