Determining biomarkers from histopathology slide images
A generalizable and interpretable deep learning model for predicting biomarker status and biomarker metrics from histopathology slide images is provided.
1. A computer-implemented method of identifying biomarkers in a digital image of a hematoxylin and eosin (H&E) stained slide of target tissue, the method comprising:
receiving the digital image to an image-based biomarker prediction system having one or more processors;
performing an image tiling process, using the one or more processors, on the digital image by separating the digital image into a plurality of tile images, where each of the plurality of tile images contains a different portion of the digital image;
applying, using the one or more processors, the plurality of tile images to a multiscale deep learning framework comprising one or more trained deep learning multiscale classifier models, each trained deep learning multiscale classifier models being trained to classify a different tissue classification for each tile image and determining a tissue classification for each of the plurality of tile images, using the multiscale deep learning framework;
identifying, using the one or more processors, cells within each of the plurality of tile images using a trained cell segmentation model; and
identifying a predicted presence of one or more biomarkers associated with the digital image based at least in part on analyzing one or more tile images adjacent to the respective tile image and analyzing a larger tile image comprising a subset of the plurality of tile images, the subset containing the one or more tile images adjacent to the respective tile image,
wherein determining the tissue classification for each of the plurality of tile images comprises analyzing one or more tile images adjacent to the respective tile image and analyzing the larger tile image comprising the subset of the plurality of tile images and producing a 3D array of values that each represent the probability that a tissue class classification label matches a tissue classification in each tile image of the subset of the plurality of tile images.
2. The method of claim 1 , wherein identifying the predicted presence of the one or more biomarkers associated with the digital image based at least in part on analyzing the one or more tile images adjacent to the respective tile image and analyzing the larger tile image comprising the subset of the plurality of tile images comprises:
identifying the predicted presence of the one or more biomarkers based at least in part on the identified cells within each of the subset of adjacent tile images having the same tissue classification.
3. The method of claim 1 , wherein identifying the predicted presence of the one or more biomarkers associated with the digital image based at least in part on analyzing the one or more tile images adjacent to the respective tile image and analyzing the larger tile image comprising the subset of the plurality of tile images comprises:
identifying the predicted presence of the one or more biomarkers based at least in part on the identified cells within each of the subset of adjacent tile images having different tissue classifications.
4. The method of claim 1 , wherein separating the digital image into a plurality of tile images comprises:
performing an image tiling process, using the one or more processors, by applying a tiling mask to the digital image to separate the digital image into the plurality of tile images.
5. The method of claim 4 , wherein the tiling mask comprises tiles of the same size.
6. The method of claim 5 , wherein the tiling mask comprises tiles having a rectangular shape.
7. The method of claim 1 , further comprising training the one or more trained deep learning multiscale classifier models by:
receiving, at the multiscale deep learning framework, a plurality of H&E slide training images from a training images dataset, each H&E slide training image having a label corresponding to a biomarker to be trained;
performing tile-based tissue classification analysis on each of the H&E slide training images;
performing a pixel-based cell segmentation analysis on each of the H&E slide training images;
optionally performing a tile-based biomarker classification analysis on each of the H&E slide training images; and
in response, generating the one or more trained deep learning multiscale classifier models.
8. The method of claim 7 , wherein each H&E slide training image comprises a plurality of tile images each having a tile-level label.
9. The method of claim 7 , further comprising, for each H&E slide training image, imputing a tile-level label for each of a plurality of tile images of the H&E slide training image.
10. The method of claim 7 , further comprising:
for each H&E slide training image, performing a tile selection process that infers a class status for each tile image in the H&E slide training image; and
based on inferred class status, discarding tile images not corresponding to a desired class, before performing the tile-based tissue classification analysis on each of the H&E slide training images, such that the tile-based tissue classification analysis is performed on only selected tile images of the H&E slide training image.
11. The method of claim 1 , wherein the one of the one or more trained deep learning multiscale classifier models are each configured as a tile-resolution Fully Convolutional Network (FCN) classification model.
12. The method of claim 11 , wherein the one or more trained deep learning multiscale classifier models are each three-class models trained to classify a different tissue interior, tissue exterior, and tissue boundary within the tile image.
13. The method of claim 1 , wherein identifying cells within the digital image tile using the trained cell segmentation model comprises:
applying, using the one or more processors, each of the plurality of tile images to the cell segmentation model and, for each tile, assigning a cell classification to one or more pixels within the tile image.
14. The method of claim 13 , wherein assigning the cell classification to one or more pixels within the tile image comprises:
identifying, using the one or more processors, the one or more pixels as a cell interior, a cell border, or a cell exterior and classifying the one or more pixels as the cell interior, the cell border, or the cell exterior.
15. The method of claim 1 , wherein the trained cell segmentation model is trained to identify and segment one or more of tumor positive cells, tumor negative cells, lymphocyte positive cells, lymphocyte negative cells, and immune cells.
16. The method of claim 1 , wherein the trained cell segmentation model is a pixel-resolution three-dimensional UNet classification model trained to classify a cell interior, a cell border, and a cell exterior.
17. The method of claim 1 , wherein the one or more biomarkers are selected from the group consisting of tumor-infiltrating lymphocytes (TILs), nucleus-to-cytoplasm (NC) ratio, ploidy, signet ring morphology, and programmed death-ligand 1 (PD-L1).
18. The method of claim 1 , wherein the different tissue classifications comprise one or more of tumor, stroma, normal, lymphocyte, fat, muscle, blood vessel, immune cluster, necrosis, hyperplasia/dysplasia, and red blood cells.
19. The method of claim 1 , wherein the different tissue classifications comprise tissue classes or cell types that are positive or negative for an Immunohistochemistry (IHC) stain target molecule.
20. The method of claim 1 , wherein the different tissue classifications comprise one or more of tumor positive, tumor negative, lymphocyte positive, and lymphocyte negative.
21. The method of claim 1 , wherein determining the tissue classification for each of the plurality of tile images comprises analyzing a larger tile image containing the one or more tile images adjacent to the respective tile image and containing the respective tile image.
22. The method of claim 21 , wherein the larger tile image is centered around the respective tile image.
23. A computing device comprising:
one or more memories; and
one or more processors configured to,
receive a digital image of a Hematoxylin and eosin (H&E) stained slide of target tissue;
perform an image tiling process on the digital image by separating the digital image into a plurality of tile images, where each of the plurality of tile images contains a different portion of the digital image;
apply the plurality of tile images to a multiscale deep learning framework comprising one or more trained deep learning multiscale classifier models, each trained deep learning multiscale classifier models being trained to classify a different tissue classification for each tile image and determine a tissue classification for each of the plurality of tile images, using the multiscale deep learning framework;
identify cells within each of the plurality of tile images using a trained cell segmentation model; and
identify a predicted presence of one or more biomarkers associated with the digital image based at least in part on analysis of one or more tile images adjacent to the respective tile image and analysis of a larger tile image comprising a subset of the plurality of tile images, the subset containing the one or more tile images adjacent to the respective tile image,
wherein determining the tissue classification for each of the plurality of tile images comprises analyzing one or more tile images adjacent to the respective tile image and analyzing the larger tile image comprising the subset of the plurality of tile images and producing a 3D array of values that each represent the probability that a tissue class classification label matches a tissue classification in each tile image of the subset of the plurality of tile images.
24. The computing device of claim 23 , wherein the one or more processors are configured to:
identify the predicted presence of the one or more biomarkers based at least in part on the identified cells within each of the subset of adjacent tile images having the same tissue classification; or
identify the predicted presence of the one or more biomarkers based at least in part on the identified cells within each of the subset of adjacent tile images having different tissue classifications.
25. The computing device of claim 23 , wherein the one or more processors are configured to:
determine the tissue classification for each of the plurality of tile images by analyzing one or more tile images adjacent to the respective tile image; or
determine the tissue classification for each of the plurality of tile images by analyzing a larger tile image containing the one or more tile images adjacent to the respective tile image and containing the respective tile image.
26. A system comprising:
a network interface to connect the system a remote computing device over a network;
one or more memories; and
one or more processors configured to,
receive a digital image of a Hematoxylin and eosin (H&E) stained slide of target tissue;
perform an image tiling process on the digital image by separating the digital image into a plurality of tile images, where each of the plurality of tile images contains a different portion of the digital image;
apply the plurality of tile images to a multiscale deep learning framework comprising one or more trained deep learning multiscale classifier models, each trained deep learning multiscale classifier models being trained to classify a different tissue classification for each tile image and determine a tissue classification for each of the plurality of tile images, using the multiscale deep learning framework;
identify cells within each of the plurality of tile images using a trained cell segmentation model;
identify a predicted presence of one or more biomarkers associated with the digital image based at least in part on analyzing one or more tile images adjacent to a respective tile image and analyzing a larger tile image comprising a subset of the plurality of tile images, the subset containing the one or more tile images adjacent to the respective tile image,
wherein determining the tissue classification for each of the plurality of tile images comprises analyzing one or more tile images adjacent to the respective tile image and analyzing the larger tile image comprising the subset of the plurality of tile images and producing a 3D array of values that each represent the probability that a tissue class classification label matches a tissue classification in each tile image of the subset of the plurality of tile images; and
generate a viewer application (App), accessible by the remote computing device over the network, the viewer App allowing a user at the remote computing device to interact, through a graphical user interface, with a report containing the predicted presence of the one or more biomarkers.
27. The system of claim 26 , wherein the one or more processors are configured to:
identify the predicted presence of the one or more biomarkers based at least in part on the identified cells within each of the subset of adjacent tile images having the same tissue classification; or
identify the predicted presence of the one or more biomarkers based at least in part on the identified cells within each of the subset of adjacent tile images having different tissue classifications.
28. The system of claim 26 , wherein the one or more processors are configured to:
allow the remote computing device to download the viewer App over the network for interacting with the report.