Systems and methods for training a statistical model to predict tissue characteristics for a pathology image

Inventors

Beck, Andrew H. • Khosla, Aditya

Assignees

PathAI Inc

Abstract

In some aspects, the described systems and methods provide for a method for training a statistical model to predict tissue characteristics for a pathology image. The method includes accessing annotated pathology images. Each of the images includes an annotation describing a tissue characteristic category for a portion of the image. A set of training patches and a corresponding set of annotations are defined using an annotated pathology image. Each of the training patches in the set includes values obtained from a respective subset of pixels in the annotated pathology image and is associated with a corresponding patch annotation determined based on an annotation associated with the respective subset of pixels. The statistical model is trained based on the set of training patches and the corresponding set of patch annotations. The trained statistical model is stored on at least one storage device.

Core Innovation

A method and corresponding system determine tissue and cell characteristics for a pathology image of a patient by processing the pathology image using a trained deep learning model trained on a plurality of annotated images. The method extracts values for one or more features based on the tissue and cell characteristics, and the extracted feature values are used to predict an entity of interest associated with the patient, where the entity of interest includes one or more of patient response, tumor molecular characteristics, and patient clinical outcomes.

The approach uses annotated images and supports tissue- and cell-focused determination through tissue and cell characteristics processing. In particular, the disclosed pipeline can include using a first deep learning model and a second deep learning model to identify cell characteristics and tissue characteristics, and the extracted one or more feature values may include spatial features for cells, including spatial distribution of cells, cellular heterogeneity, and texture.

The predicted entity of interest is output for storing on at least one storage device, and the disclosure also includes output behaviors such as displaying annotations overlaid on a pathology image describing determined tissue and cell characteristics. The entity of interest may further include tumor molecular characteristics including genomic markers, and may be predicted using clinical metadata associated with the patient. The disclosed embodiments are implemented as a method, a system, and a non-transitory computer-readable medium containing instructions executed by at least one processor.

Claims Coverage

The independent claims are directed to predicting an entity of interest from a patient pathology image using trained deep learning on annotated images, with feature extraction based on tissue and cell characteristics and prediction output for storage. Three independent claims are present, differing in claim category as method, system, and non-transitory computer-readable medium while retaining the same core prediction pipeline.

Across the independent claims, the core inventive coverage is the use of a trained deep learning model on annotated pathology images to determine tissue and cell characteristics, extract feature values including spatial features, and predict an entity of interest for the patient that includes patient response, tumor molecular characteristics including genomic markers, and/or patient clinical outcomes, with implementations as a method, system, and non-transitory computer-readable medium and with optional refinements such as separate cell and tissue models, clinical metadata, and annotation overlay output.

Stated Advantages

Documented Applications