Tissue microenvironment analysis based on tiered classification and clustering analysis of digital pathology images
Segmentation or other classification of digital pathology images with a deep learning model allows for sophisticated spatial features for cancer diagnosis to be extracted in an automated, fast, and accurate manner. A tiered analysis of tissue structure based in part on deep learning methods is provided. First, tissues depicted in a digital pathology image are segmented into cellular compartments (e.g., epithelial and stromal compartments). Second, the heterogeneity in the different cellular compartments are examined based on a clustering algorithm. Tissue can then be characterized in terms of inertia (or other spatial measures or features), which can be used to recognize disease. In some instances, multidimensional inertia (i.e., inertia computed in different cellular compartments or clustered components) can be used as an indicator of disease and its outcome.
1. A method for characterizing a tissue based on analysis of digital pathology images of the tissue, the method comprising:
(a) accessing segmented digital pathology image data with a computer system, wherein the segmented digital pathology image data comprise digital pathology images of a tissue that have been segmented into two or more different cellular compartments of the tissue;
(b) applying the segmented digital pathology image data to a clustering algorithm using the computer system, generating output as clustered feature data comprising clusters of the segmented digital pathology image data that indicate separated tissue microenvironment components of the tissue;
(c) computing spatial feature parameter data for clusters in the clustered feature data using the computer system; and
(d) generating a report with the computer system, wherein the report characterizes the tissue based on the spatial feature parameter data, wherein the report indicates a likelihood of disease in the tissue.
2. The method of claim 1 , wherein the two or more different cellular components include epithelial and stromal compartments.
3. The method of claim 1 , wherein the tissue microenvironment components include at least one of immune cells and cancer associated with fibroblasts.
4. The method of claim 1 , wherein the clustering algorithm is an unsupervised clustering algorithm.
5. The method of claim 4 , wherein the unsupervised clustering algorithm is a K-mean clustering algorithm.
6. The method of claim 5 , wherein the K-mean clustering algorithm uses a number of clusters ranging from two to five clusters.
7. The method of claim 1 , wherein the tissue is breast tissue.
8. The method of claim 1 , wherein accessing the segmented digital pathology image data with the computer system comprises:
accessing digital pathology image data with the computer system, wherein the digital pathology image data comprise digital pathology images of the tissue; and
segmenting the digital pathology image data with the computer system, generating output as the segmented digital pathology image data.
9. The method of claim 8 , wherein segmenting the digital pathology image data with the computer system comprises:
accessing a deep learning model with the computer system, wherein the deep learning model has been trained to segment digital pathology images into different cellular components; and
applying the digital pathology image data to the deep learning model using the computer system, generating output as the segmented digital pathology image data.
10. The method of claim 9 , wherein the deep learning model is a convolutional neural network (CNN) model.
11. The method of claim 10 , wherein the CNN model has been trained on hematoxylin and eosin (H&E) stained images.
12. The method of claim 11 , wherein the H&E stained images used to train the CNN model comprise H&E images obtained from specimens including normal breast tissue, usual hyperplasia, atypical hyperplasia with columnar change, and atypical hyperplasia without columnar change, and ductal carcinoma in-situ.
13. The method of claim 1 , wherein computing the spatial feature parameter data for clusters in the clustered feature data comprise computing inertia for clusters in the clustered feature data.
14. The method of claim 13 , wherein the report characterizes the tissue based on the inertia computed for the clusters in the clustered feature data.
15. The method of claim 14 , wherein the two or more different cellular components include epithelial and stromal compartments, and the report characterizes the tissue based on the inertia computed for clusters in at least one of the epithelial compartment and the stromal compartment.
16. The method of claim 15 , wherein the report characterizes the tissue based on the inertia computed for both the epithelial compartment and the stromal compartment.
17. The method of claim 1 , wherein the report identifies spatial characteristics of early stage disease that are indicative of triaging biopsies from the tissue.