Method for determining biomarkers from pathological tissue slide images

Abstract

Providing novel technologies that enable easy diagnosis of TIL, PD-L1, and other biomarkers using H&E imaging to efficiently identify and characterize biomarkers across population groups, create more optimized drug therapy recommendations and protocols, and improve prediction of disease progression. [Solution] A generalizable and interpretable deep learning model is provided for predicting biomarker status and biomarker metrics from pathological tissue slide images.

Claims

1. A computer-implemented method for identifying biomarkers in digital images of hematoxylin and eosin (H&E) stained slides of target tissue, wherein the method is: Receiving digital images in an image-based biomarker prediction system having one or more processors. Using one or more processors, a digital image is separated into multiple tile images using image tiling technology, where each of the multiple tile images contains a different portion of the digital image. Applying multiple tile images to a deep learning framework containing one or more trained biomarker classification models using one or more processors, where each trained biomarker classification model is trained to classify different biomarkers, where the deep learning framework includes a single-scale deep learning framework or a multi-scale deep learning framework; Using one or more processors, predict the biomarker classification for each of multiple tile images using one or more trained biomarker classification models. Here, one or more biomarker classification models are trained using molecular training datasets that correspond to multiple training tissue samples, each containing molecular data based on sequencing of substantially similar samples associated with each training tissue sample, and containing multiple subsets of molecular data clustered by biomarker; To determine the predicted presence of one or more biomarkers in the target tissue from the predicted biomarker classification of each tile image; and The method comprising generating a report that includes a digital image and a digital overlay that visualizes the predicted presence of one or more biomarkers.

2. The above method further, The computer implementation method according to claim 1, comprising receiving digital images from a pathology slide scanner system via an electrical network.

3. The method further involves receiving molecular training datasets for multiple training tissue samples, wherein the molecular training datasets include RNA transcriptome counts from sequencing of substantially the same samples associated with each training tissue sample; Clustering of molecular training datasets is performed to identify one or more molecular data subsets corresponding to different biomarkers. For each of one or more subsets of molecular data, multiple digital images of H&E-stained training slides of training tissue samples corresponding to each biomarker are received by an image-based biomarker prediction system having one or more processors, and Applying the multiple tile images to a deep learning framework that includes multiple digital images of the H&E stained training slides and one or more biomarker classification models trained to classify different biomarkers based on the molecular training dataset. A computer implementation method according to claim 1 or 2, including the method described in claim 1 or 2.

4. This includes performing multiple instance learning processes on multiple digital images of the H&E stained training slides, wherein the multiple instance learning processes are as follows: A tile selection process is provided that assigns a class status to each tile image in the H&E slide training image. Based on the assigned class status, discard a subset of the tile images before applying the remaining multiple tile images to the deep learning framework. The method according to claim 3, including the method described in claim 3.

5. The method according to claim 3, wherein each of the multiple digital images of the H&E-stained training slide of the training tissue sample has a slide level label.

6. The method according to claim 3, wherein each of the multiple digital images of the H&E-stained training slide of the training tissue sample is unlabeled.

7. The method according to claim 1, wherein the single-scale deep learning framework is a convolutional neural network having a ResNet configuration or an initiation configuration.

8. The method according to claim 1, wherein the one or more biomarkers are selected from the group consisting of consensus molecular subtypes (CMS) and homologous recombination deficiencies ("HRD").

9. The method according to claim 1, wherein the digital overlay includes an overlay element for identifying the tumor content or tumor percentage of a digital image.

Images