Methods and systems for training a model to diagnose abnormalittes in tissue samples
AssigneeImagene AI LTD.
InventorsYehonatan Zalah, Dean Bitan, Shahar Porat
Life of this patent
Estimating the term…
Abstract
The present invention relates to artificial intelligence and machine learning methods and systems for diagnosing a disease, disorder or pathological condition. More particularly, the invention relates to methods and systems for training a machine learning model to diagnose a disease, disorder or pathological condition in a biopsy sample, to detect tissue abnormality in a biopsy sample, as well as to detect gene mutations expressed in the cells that are present in the biopsy sample.
Claims27
A method for training a machine learning model to detect the presence of a gene mutation or a tissue abnormality in a biopsy sample, the method comprising the steps of:
- receiving a plurality of stained and annotated histology images, wherein each of the plurality of stained and annotated histology images includes at least one annotation of a tissue characteristic in at least a portion of the image;
- selecting slide patches out of maximum overlapping slide patches of each of the plurality of histology images to be used as training patches for the machine learning model based on a total score determined for each of the maximum overlapping slide patches, wherein the total score is indicative of the relevance of the patch as a training patch based on the number of pixels in the patch that are positive for the presence of a tissue characteristic, the type of said tissue characteristic, and the number of pixels in the patch that are positive for tissue abnormality according to the received annotations;
- training the machine learning model by receiving the slide patches selected as training patches;
- and processing the selected slide patches, based on the annotations of the tissue characteristics included in the histology image using at least one sub-model configured to predict the presence of a gene mutation or a tissue abnormality in each of said histology images, thereby obtaining at least one prediction indicating a probability that a gene mutation or a tissue abnormality is present in the histology image;
- and
- validating the accuracy of the at least one prediction.
A method for increasing the amount of training data provided to a machine learning model configured to detect the presence of a gene mutation or tissue abnormality, the method comprising the steps of:
- receiving a plurality of stained and annotated histology images, wherein the stained and annotated histology images are prepared from biopsy samples harvested from a group of patients who were diagnosed with a disease, disorder or pathological condition or who were detected to carry a gene mutation, and wherein each of the plurality of stained and annotated histology images includes at least one annotation of a tissue characteristic in at least a portion of the image;
- and
- selecting slide patches out of maximum overlapping slide patches of the plurality of histology images to be used as training patches for the machine learning model based on a total score determined for each of the maximum overlapping slide patches, wherein the total score is indicative of the relevance of the patch as a training patch based on the number of pixels in the patch that are positive for the presence of a tissue characteristic, the type of said tissue characteristic, and the number of pixels in the patch that are positive for tissue abnormality according to the received annotations.
Dependencies are inferred from the claim text.