Predicting prostate cancer biochemical recurrence and metastasis with computer extracted features from tumor nuclei and benign regions

Patent metadata

Field	Value
Publication number	US-10776607-B2
Application number	US-201815983397-A
Country	US
Kind code	B2
Filing date	May 18, 2018
Priority date	May 19, 2017
Publication date	Sep 15, 2020
Grant date	Sep 15, 2020

Abstract

Embodiments predict biochemical recurrence (BCR) or metastasis by accessing a set of images of a region of tissue demonstrating cancerous pathology, including a tumor region and a tumor adjacent benign (TAB) region, the set of images including a first stain type image, and a second stain type image; segmenting cellular nuclei represented in the first and second image; generating a combined feature set by extracting at least one feature from each of a tumor region and TAB region represented in the first image, and a tumor region and TAB region represented in the second image, providing the combined feature set to a machine learning classifier; receiving, from the classifier, a probability that the region of tissue will experience BCR or metastasis; and generating a classification of the region of tissue as likely to experience BCR or metastasis, or unlikely to experience BCR or metastasis.

First claim

What is claimed is:

1. A non-transitory computer-readable storage device storing computer executable instructions that when executed control a processor to perform operations, the operations including:
accessing a set of images of a region of tissue demonstrating prostate cancer (PCa), where an image has a plurality of pixels, a pixel having an intensity, where the region of tissue includes a tumor region and a tumor adjacent benign (TAB) region, and where the set of images includes a digitized hematoxylin and eosin (H&E) stained image of the region of tissue, and a Feulgen stained image of the region of tissue;
segmenting cellular nuclei represented in the H&E stained image and the Feulgen stained image using a watershed approach;
extracting a first set of radiomic features from the H&E stained image based, at least in part, on the segmented nuclei;
extracting a second set of radiomic features from the Feulgen stained image based, at least in part, on the segmented nuclei;
generating a combined feature set from the first set of radiomic features and the second set of radiomic features, where the combined feature set includes at least one feature extracted from the tumor region represented in the H&E stained image and at least one feature extracted from the TAB region represented in the H&E stained image, and at least one feature extracted from the tumor region represented in the Feulgen stained image and at least one feature extracted from the TAB region represented in the Feulgen stained image, wherein the combined feature set includes ten features;
providing the combined feature set to a machine learning classifier;
receiving, from the machine learning classifier, a probability that the region of tissue will experience biochemical recurrence (BCR), where the probability is based, at least in part, on the combined feature set; and
generating a classification of the region of tissue as likely to experience BCR or unlikely to experience BCR based, at least in part, on the probability.

2. The non-transitory computer-readable storage device of claim 1 , the operations further including:
receiving, from the machine learning classifier, a second probability that the region of tissue will experience metastasis, where the second probability is based, at least in part, on the combined feature set; and
generating a classification of the region of tissue as likely to experience metastasis or unlikely to experience metastasis based, at least in part, on the second probability.

3. The non-transitory computer-readable storage device of claim 1 , the operations further including:
generating a PCa treatment plan based, at least in part, on the classification.

4. The non-transitory computer-readable storage device of claim 1 , where the H&E stained image is a digitized image of an H&E slide of the region of tissue scanned at a resolution of 0.5 microns per pixel.

5. The non-transitory computer-readable storage device of claim 1 , where the Feulgen stained image is a digitized image of a Feulgen stained slide of the region of tissue scanned at a resolution of 0.5 microns per pixel.

6. The non-transitory computer-readable storage device of claim 1 , where the first set of radiomic features includes a global cell graph feature, a local cell graph feature, a shape feature, a cell orientation entropy (COrE) feature, and a texture feature, and where the second set of radiomic features includes a global cell graph feature, a local cell graph feature, a shape feature, a cell orientation entropy (COrE) feature, and a texture feature.

7. The non-transitory computer-readable storage device of claim 6 , where a global cell graph feature or a local cell graph feature is a Voronoi diagram, a Delaunay triangulation plot, or a minimum spanning tree.

8. The non-transitory computer-readable storage device of claim 1 , where the machine learning classifier is a Random Forest classifier.

9. The non-transitory computer-readable storage device of claim 1 , the operations further including training the machine learning classifier.

10. An apparatus for predicting prostate cancer (PCa) biochemical recurrence (BCR), the apparatus comprising:
a processor;
a memory configured to store a set of digitized images of a region of tissue demonstrating PCa, where a member of set of digitized images has a plurality of pixels, a pixel having an intensity, where the region of tissue includes a tumor region and a tumor adjacent benign (TAB) region;
an input/output (I/O) interface;
a set of circuits comprising an image acquisition circuit, a nuclei segmentation circuit, a feature extraction circuit, and a combined classification circuit; and
an interface that connects the processor, the memory, the I/O interface, and the set of circuits;
the image acquisition circuit configured to access the set of digitized images, where the set of digitized images includes a hematoxylin and eosin (H&E) stained image of the region of tissue, and a Feulgen stained image of the region of tissue;
the nuclei segmentation circuit configured to segment cellular nuclei represented in the H&E stained image and the Feulgen stained image using a watershed approach;
the feature extraction circuit configured to:
extract an H&E set of radiomic features from the tumor region and the TAB region of the H&E stained image based, at least in part, on the segmented nuclei;
extract a Feulgen set of radiomic features from the tumor region and the TAB region of the Feulgen stained image based, at least in part, on the segmented nuclei;
generate a combined feature set from the H&E set and the Feulgen set, where the combined feature set includes at least one feature extracted from each of the tumor region and the TAB region of the H&E stained image, and at least one feature extracted from each of the tumor region and the TAB region of the Feulgen stained image, wherein the combined feature set includes ten features; and
provide the combined feature set to the combined classification circuit;

the combined classification circuit configured to:
receive the combined feature set from the feature extraction circuit;
compute a probability that the region of tissue will experience BCR based, at least in part, on the combined feature set; and
generate a classification of the region of tissue as likely to experience BCR or unlikely to experience BCR based, at least in part, on the probability.

11. The apparatus of claim 10 , where the combined feature set includes a global cell graph feature, a local cell graph feature, a shape feature, a cell orientation entropy (COrE) feature, and a texture feature.

12. The apparatus of claim 10 , where the combined classification circuit is further configured to:
receive the combined feature set from the feature extraction circuit;
compute a second probability that the region of tissue will experience metastasis based, at least in part, on the combined feature set; and
generate a second classification of the region of tissue as likely to experience metastasis or unlikely to experience metastasis based, at least in part, on the second probability.

13. The apparatus of claim 12 , where the combined classification circuit is a machine learning classifier configured to compute the probability or the second probability based, at least in part, on the combined feature set using a random forest machine learning approach.

14. The apparatus of claim 12 , the set of circuits further comprising:
a treatment plan generation circuit configured to generate a PCa treatment plan based, at least in part, on the classification or the second classification; and
a display circuit config

Assignees

• Univ Case Western Reserve

Inventors

• Madabhushi Anant
• Gawlik Anna
• Lee George

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V20/698

  Matching; Classification · CPC title

• G16H50/30 Primary

  for calculating health indices; for individual health risk assessment · CPC title

• G01N33/57555

  of the prostate · CPC title

• G06F18/2413

  based on distances to training or reference patterns · CPC title