Methods for characterizing and treating a cancer type using cancer images
Described herein are methods, systems, devices and computer program products for characterizing or identifying a type of cancer. Also described are methods of treating a characterized or identified chancer. For example, certain methods may be used to characterize a homologous recombination deficiency status of a cancer.
1 . A computer-implemented method of identifying a cancer as homologous recombination deficient or homologous recombination proficient, comprising:
inputting image data comprising tiles or pixels corresponding to a digitized image of the cancer into a trained machine learning model, trained on training image data comprising a plurality of digitized images associated with a homologous recombination status;
processing, using one or more processors, the inputted image data by the trained machine learning model that associates the inputted image data with a homologous recombination deficiency status, wherein the trained machine learning model comprises (i) a deep learning model (ii) a convolutional neural network (CNN) learning model, (iii) an ensemble model or (iv) an image classification model, and wherein the trained machine learning model comprises one or more weighted features including a first feature comprising a mutational signature score associated with the cancer and a second feature comprising a large-scale state transition (LST) score associated with the cancer;
outputting, from the trained machine learning model, an output indicating the homologous recombination deficiency status of the cancer; and
identifying the cancer as homologous recombination deficient (HRD) or homologous recombination proficient (HRP) based on the output indicating the homologous recombination deficiency status of the cancer.
2 . The computer-implemented method of claim 1 , wherein the homologous recombination deficiency status comprises (i) a likelihood that the cancer is homologous recombination deficient or a likelihood that the cancer is homologous recombination proficient and/or (ii) a binary determination that the cancer is homologous recombination deficient or homologous recombination proficient.
3 . The computer-implemented method of claim 1 , wherein the image data is obtained from a stained image of cancer or an image of a specimen sample of cancer or a histology image of cancer or any combination of the foregoing.
4 . The computer-implemented method of claim 1 , wherein the trained machine learning (ML) model is an image classification model that is a binary image classification model configured to classify the cancer in the inputted cancer image according to classes which respectively represent HRD and HRP.
5 . A computer-implemented method of diagnosing cancer as HRD in a human comprising the steps of applying the method of claim 1 to an image of the cancer.
6 . The computer-implemented method of claim 1 , wherein the cancer is selected from the group consisting of epithelial cancer, breast cancer, ovarian cancer, fallopian tube cancer, peritoneal cancer, colon cancer, colorectal cancer, rectal cancer, cholagiocarcinoma, gastric cancer, and endometrial cancer.
7 . A computer-implemented method for identifying whether a cancer in a subject is homologous recombination deficient (HRD) comprising the steps of:
i. receiving an image of the subject's cancer, wherein the image comprises tiles or pixels corresponding to a digitized image of the cancer, and
ii. processing said image, using one or more processors, with an image analysis system and using a trained image classification model to determine a homologous recombination deficiency status to identify therefrom if the cancer is HRD, wherein the trained image classification model comprises one or more weighted features including a first feature comprising a mutational signature score associated with the cancer and a second feature comprising a large-scale state transition (LST) score associated with the cancer, and
iii. outputting a result including the HRD status from the image processing step.
8 . The computer-implemented method of claim 7 , wherein the image classifier is a binary image classifier for classes for HRD and not-HRD.
9 . A system to identify a homologous recombination (HR) status of a cancer of a subject, preferably whether the cancer of the subject is HR deficient (HRD) or whether the cancer is HRD or other than HRD, the system comprising an image analysis system and one or more processors configured to:
i) receive as input an image of the cancer of the subject, wherein the image comprises tiles or pixels corresponding to a digitized image of the cancer,
ii) process the cancer image with a trained machine learning (ML) model to determine the HR status of the cancer of the subject from the image of the cancer, wherein the trained ML model is (a) trained by supervised learning, (b) a deep learning (DL) model, (c) configured by at least one artificial neural network, (d) a convolutional neural network learning model, (e) an ensemble model, or (f) any combination of the foregoing options (a) to (e), wherein the trained machine learning model comprises an image classification model and one or more weighted features including a first feature comprising a mutational signature score associated with the cancer and a second feature comprising a large-scale state transition (LST) score associated with the cancer, and
iii) provide an output to indicate the HR status for the cancer based on the processing of the image of said cancer by the trained ML model.
10 . The system of claim 9 , wherein the image analysis system has an image pre-processor to process the image of the cancer received as input prior to processing by the trained ML model.
11 . The system of claim 9 , wherein the image is of a specimen sample of the cancer or is of a stained cancer sample or is in the form of image data or is any combination of the foregoing.
12 . The system of claim 9 , wherein the trained ML model comprises a classifier to classify the cancer in the input cancer image with respect to a class set which is representative of a set of HR classes, preferably a class for HRD and a class for one or more other HR classes.
13 . The system of claim 12 , wherein the output is the classification of the cancer in the input cancer image with respect to the class set or the classifier is a binary classifier for classes representing HRD and HRP.
14 . The system of claim 9 , wherein the cancer is selected from the group consisting of epithelial cancer, breast cancer, ovarian cancer, fallopian tube cancer, peritoneal cancer, colon cancer, colorectal cancer, rectal cancer, cholagiocarcinoma, gastric cancer, and endometrial cancer.
15 . The system of claim 9 , wherein the system outputs a diagnosis of cancer as HRD in a human based on processing the image of the cancer.
16 . A computer-implemented method performed by a data processing system comprising one or more processors for identifying a cancer as homologous recombination deficient (HRD) or homologous recombination proficient (HRP), comprising:
inputting image data comprising tiles or pixels corresponding to a digitized image of the cancer into the data processing system;
accessing a computer program that includes a trained machine learning model that associates the inputted image data with a homologous recombination deficiency status, wherein the trained machine learning model comprises one or more weighted features including a first feature comprising a mutational signature score associated with the cancer and a second feature comprising a large-scale state transition (LST) score associated with the cancer;
executing the computer program with the trained machine learning model to process the inputted image data to identify image data features;
determining, based on the identified image data features, the homologous recombination deficiency status for the inputted image data; and
storing, in memory, a data structure with fields representing the inputted image data and its homologous recombination deficiency status.
17 . A computer program product comprising a computer readable storage medium comprising instructions which, when the program is executed by a computer or data processor(s), cause the computer or data processor(s) to process a digitized image of a cancer to determine or predict a homologous recombination deficient (HRD) status of the cancer, wherein the instructions when executed implement (a) a trained machine learning model, wherein the trained machine learning model comprises an image classification model and one or more weighted features including a first feature comprising a mutational signature score associated with the cancer and a second feature comprising a large-scale state transition (LST) score associated with the cancer, and wherein the instructions when executed provide an output with an indication of the HRD status.
18 . A computer-readable data carrier having stored thereon the computer program product of claim 17 .