Granted Patent B2

US 12,524,826 · App. 17/684,191 · Granted Jan 13, 2026

Inventors: Stephen Yip (Chicago, IL); Irvin Ho (Wilmette, IL); Lingdao Sha (Chicago, IL); Boleslaw Osinski (Chicago, IL); Aly Azeem Khan (Chicago, IL); Andrew J. Kruger (Chicago, IL); Michael Carlson (Chicago, IL); Abel Greenwald (Chicago, IL); Caleb Willis (Chicago, IL)

Assignee: TEMPUS AI, INC.

G06T1/20G06F18/21G06F18/2431G06T7/0012G06T7/11G06T11/00G06V10/44G06V10/764G06V10/82G06T2207/20081G06T2207/30024G06T2207/30096

Abstract

Claims (44)

1 . A computer-implemented method of identifying one or more molecular biomarkers in a digital image of a hematoxylin and eosin (H&E) stained slide, the computer-implemented method comprising:

receiving, via one or more processors, the digital image of the H&E stained slide at an image-based biomarker prediction system;

separating, via one or more processors, the digital image of the H&E stained slide into a plurality of tile images, where each of the plurality of tile images corresponds to a respective region of a tissue in the digital image;

applying the plurality of tile images to one or more trained classifiers trained on information including both

(1) a first plurality of training images corresponding to training H&E stained slides,

wherein each training H&E stained slide is separated into an initial plurality of training H&E stained tiles,

wherein each training H&E stained tile is associated with an inferred tissue classification status, and

wherein the first plurality of training images comprises training H&E stained tiles, from the initial plurality of training H&E stained tiles, having an inferred tissue classification status corresponding to a desired tissue classification, and

(2) a second plurality of training images corresponding to training immunohistochemistry (IHC) stained slides,

wherein each of training IHC stained slides is separated into training IHC stained tiles, and

wherein each of the training IHC stained slides are stained for the one or more molecular biomarkers and;

predicting, via the one or more trained classifiers, one or more molecular biomarkers for each of the plurality of tile images having a target tissue; and

identifying, via the one or more processors, the one or more molecular biomarkers in the digital image based at least in part on the one or more molecular biomarkers of the plurality of tile images having the target tissue.

2 . The computer-implemented method of claim 1 , wherein one or more molecular biomarkers include programmed death-ligand 1 (PD-L1), phosphatase and tensin homolog (PTEN), epidermal growth factor receptor (EGFR), β catenin/catenin β1, neurotrophic tyrosine kinase receptor (NTRK), phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit a (PIK3CA), receptor tyrosine-protein kinase erbB-2 (HER2), androgen receptor (AR), estrogen receptor (ER), or progesterone receptor (PR).

3 . The computer-implemented method of claim 1 , wherein the target tissue is tumor tissue.

4 . The computer-implemented method of claim 3 , wherein the inferred tissue classification status include stroma, epithelium, lymphocyte, or necrosis.

5 . The computer-implemented method of claim 3 , wherein the inferred tissue classification status include tissues other than tumor tissue.

6 . The computer-implemented method of claim 3 , wherein the inferred tissue classification status includes not tissue.

7 . The computer-implemented method of claim 1 , wherein the target tissue is tumor tissue having lymphocytes.

8 . The computer-implemented method of claim 1 , wherein separating the digital image into the 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.

9 . The computer-implemented method of claim 8 , wherein the tiling mask comprises tiles of an equivalent size.

10 . The computer-implemented method of claim 8 , wherein the tiling mask comprises tiles having a rectangular shape.

11 . The computer-implemented method of claim 8 , wherein the tiling mask comprises tiles characterized by topology and/or morphology of pixels or groups of pixels.

12 . The computer-implemented method of claim 1 , wherein the image-based biomarker prediction system is communicatively coupled to a pathology slide scanner system through a communication network, such that the image-based biomarker prediction system receives the digital image from the pathology slide scanner system over the communication network, or the image-based biomarker prediction system is contained within a pathology slide scanner system.

13 . A system comprising a pathology slide scanner system configured to perform the computer-implemented method of claim 1 .

14 . The computer-implemented method of claim 1 , further comprising generating a report containing:

the digital image and indicating the one or more molecular biomarkers in the digital image,

the digital image and a digital overlay visualizing the one or more molecular biomarkers in the digital image,

the digital image and a digital overlay visualizing regions of tissue having the one or more molecular biomarkers in the digital image, or

one or more clinical trials or therapies associated with the one or more molecular biomarkers in the digital image.

15 . The computer-implemented method of claim 1 further comprising:

separating, via the one or more processors, tiles of the plurality of tile images having a target tissue from tile images of the plurality of tile images having one or more other targets by

applying the plurality of tile images to a deep learning framework having one or more trained tissue classifiers, wherein the one or more trained tissue classifiers are each trained to classify the target tissue or each of the one or more other targets.

16 . The computer-implemented method of claim 15 , further comprising: identifying, using the one or more processors, cells within the digital image using a trained cell segmentation model within the deep learning framework.

17 . The computer-implemented method of claim 16 , further comprising separating tiles of the plurality of tile images having the target tissue based on a target tissue classification or an other target classification determined for each tile image and on the cells within the digital image.

18 . The computer-implemented method of claim 16 , 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.

19 . The computer-implemented method of claim 16 , further comprising training the one or more trained tissue classifiers by:

receiving, at the deep learning framework, a plurality of H&E slide training images from a training images dataset, each of the plurality of H&E slide training images having a label corresponding to a tissue classification;

performing tile-based tissue classification analysis on each of the plurality of H&E slide training images;

performing a pixel-based cell segmentation analysis on each of the plurality of H&E slide training images;

performing a tile-based tissue classification analysis on each of the plurality of H&E slide training images; and

in response, generating the one or more trained tissue classifiers.

20 . The computer-implemented method of claim 1 , wherein the one or more molecular biomarkers in the digital image is communicated to a genomic sequencing system for performing molecular sequencing in response.

Assignments (4)

Continuity (10)

References Cited (334)