Patent US10650520

Systems and methods for training a statistical model to predict tissue characteristics for a pathology image
[+] Show Details

US10650520

1. A method for training a statistical model to predict tissue characteristics for a pathology image, the method comprising: (9) (4)

accessing a plurality of annotated pathology images, each of which includes at least one annotation describing one of a plurality of tissue characteristic categories for a portion of the image;

defining a set of training patches and a corresponding set of annotations using the at least one annotated pathology image, wherein each of the training patches in the set includes values obtained from a respective subset of pixels in the at least one annotated pathology image and is associated with a corresponding patch annotation determined based on an annotation associated with the respective subset of pixels;

training the statistical model based on the set of training patches and the corresponding set of patch annotations; and

storing the trained statistical model on at least one storage device.

2. The method of claim 1, further comprising: (1) (4)

processing, using the trained statistical model, at least one unannotated portion of an image of the plurality of annotated pathology images to predict an annotation of one of the plurality of tissue characteristic categories for the unannotated portion of the image;

presenting, via a user interface, the predicted annotation to a user;

receiving, via the user interface, an indication of whether the predicted annotation is accurate; and

annotating the at least one unannotated portion of the image with the predicted annotation when the indication specifies that the predicted annotation is accurate.

3. The method of claim 2, wherein when the indication specifies that the predicted annotation is not accurate, the method further comprises: (1) (4)

updating the predicted annotation based on the received indication;

redefining a set of training patches from the plurality of annotated pathology images;

retraining the statistical model based on the redefined set of training patches and the respective annotations; and

storing the retrained statistical model on the at least one storage device.

4. The method of claim 3, wherein the updated annotation has a category determined based on the received indication. (1) (0)

6. The method of claim 1, wherein defining the set of training patches from the plurality of annotated pathology images comprises: (2) (1)

9. The method of claim 1, wherein defining the set of training patches from the plurality of annotated pathology images comprises: (0) (1)

10. The method of claim 1, wherein defining the set of training patches from the plurality of annotated pathology images comprises: (0) (1)

11. The method of claim 1, wherein at least one dimension of each of the training patches in the set is larger than a corresponding dimension of input to the statistical model, wherein additional pixels in each of the training patches in the set are included when randomly resampling the respective training patch for data augmentation. (0)

12. The method of claim 1, wherein the statistical model comprises a convolutional neural network including a plurality of layers, wherein there is no padding applied to an output of any layer of the plurality of layers. (0)

13. The method of claim 1, wherein the statistical model comprises a convolutional neural network including a plurality of layers, wherein at least one of the plurality of layers, having input of size N*N, a convolution filter of size K, and a stride S, is aligned such that (N−K)/S is an integer. (0)

14. The method of claim 1, wherein the plurality of tissue characteristic categories includes one or more categories selected from a group consisting of cancer epithelium, cancer stroma, normal tissue, necrosis, lymphoid, macrophage, lymphocyte, fibroblast, plasma cell, pulmonary macrophage, melanoma, pigmented macrophage, endothelial cell, mitotic figure, nucleoli, nuclear pleomorphism, nerve, adipocyte, peri-neural invasion, epithelial and epithelial tubule formation. (0)

15. The method of claim 1, wherein the trained statistical model is a first trained statistical model, wherein the method further comprises: (1) (4)

magnifying the plurality of annotated pathology images to create a magnified plurality of pathology images;

defining a second set of training patches from the magnified plurality of annotated pathology images;

training a second statistical model based on the second set of training patches and respective annotations; and

storing the trained second statistical model on at least one storage device.

16. The method of claim 15, further comprising: (0) (1)

17. A system for training a statistical model to predict tissue characteristics for a pathology image, the system comprising: (9) (2)

at least one computer hardware processor; and

at least one non-transitory computer-readable storage medium storing processor-executable instructions that, when executed by the at least one computer hardware processor, cause the at least one computer hardware processor to perform: accessing a plurality of annotated pathology images, each of which includes at least one annotation describing one of a plurality of tissue characteristic categories for a portion of the image;

training the statistical model based on the set of training patches and the corresponding set of patch annotations; and

storing the trained statistical model on at least one storage device.

18. The system of claim 17, wherein the processor-executable instructions further cause the at least one computer hardware processor to perform: (1) (4)

presenting, via a user interface, the predicted annotation to a user;

receiving, via the user interface, an indication of whether the predicted annotation is accurate; and

annotating the at least one unannotated portion of the image with the predicted annotation when the indication specifies that the predicted annotation is accurate.

19. The system of claim 18, wherein when the indication specifies that the predicted annotation is not accurate, the processor-executable instructions further cause the at least one computer hardware processor to perform: (1) (4)

updating the predicted annotation based on the received indication;

redefining a set of training patches from the plurality of annotated pathology images;

retraining the statistical model based on the redefined set of training patches and the respective annotations; and

storing the retrained statistical model on the at least one storage device.

20. The system of claim 19, wherein the updated annotation has a category determined based on the received indication. (1) (0)

22. The system of claim 17, wherein defining the set of training patches from the plurality of annotated pathology images comprises: (2) (1)

25. The system of claim 17, wherein defining the set of training patches from the plurality of annotated pathology images comprises: (0) (1)

26. The system of claim 17, wherein defining the set of training patches from the plurality of annotated pathology images comprises: (0) (1)

27. The system of claim 17, wherein at least one dimension of each of the training patches in the set is larger than a corresponding dimension of input to the statistical model, wherein additional pixels in each of the training patches in the set are included when randomly resampling the respective training patch for data augmentation. (0)

28. The system of claim 17, wherein the statistical model comprises a convolutional neural network including a plurality of layers, wherein there is no padding applied to an output of any layer of the plurality of layers. (0)

29. The system of claim 17, wherein the statistical model comprises a convolutional neural network including a plurality of layers, wherein at least one of the plurality of layers, having input of size N*N, a convolution filter of size K, and a stride S, is aligned such that (N−K)/S is an integer. (0)

30. The system of claim 17, wherein the plurality of tissue characteristic categories includes one or more categories selected from a group consisting of cancer epithelium, cancer stroma, normal tissue, necrosis, lymphoid, macrophage, lymphocyte, fibroblast, plasma cell, pulmonary macrophage, melanoma, pigmented macrophage, endothelial cell, mitotic figure, nucleoli, nuclear pleomorphism, nerve, adipocyte, peri-neural invasion, epithelial and epithelial tubule formation. (0)

31. The system of claim 17, wherein the trained statistical model is a first trained statistical model, wherein the processor-executable instructions further cause the at least one computer hardware processor to perform: (1) (4)

magnifying the plurality of annotated pathology images to create a magnified plurality of pathology images;

defining a second set of training patches from the magnified plurality of annotated pathology images;

training a second statistical model based on the second set of training patches and respective annotations; and

storing the trained second statistical model on at least one storage device.

32. The method of claim 31, the processor-executable instructions further cause the at least one computer hardware processor to perform: (0) (1)

View Abstract and Specification Size

PDF with Images and Document Face >

Full Text Publication >

Patent US10650520

Systems and methods for training a statistical model to predict tissue characteristics for a pathology image

Patent Matrix® Search