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

The disclosed approach predicts an entity of interest for a pathology image by using at least one machine learning model to extract values for one or more features from input data representing an annotated pathology image. The annotated pathology image comprises one or more annotations for the pathology image, and the extracted feature values are processed to predict an entity of interest in which the entity of interest comprises patient clinical outcomes. Output data representing the predicted entity of interest is generated based on the model processing of the extracted feature values.

The machine learning model is trained from extracted values for the one or more features from a plurality of annotated training pathology images, where the training pathology images are collected from patients participating in a randomized controlled clinical trial. In further implementations, the training and/or prediction is refined by incorporating additional patient data modalities such as clinical metadata values and genomic data, transcriptomic data, and/or protein expression data. The target outcome can include survival time as part of predicting patient clinical outcomes.

The feature values are derived from annotated pathology image representations including, in some implementations, a heat map, where values are extracted by applying one or more functions to the input data and/or to the heat map. The features can encompass quantitative pathology features and spatial relationships, including tissue-category features and metrics such as areas, cellular statistics, nuclear grading metrics, and spatial distance features between cell types and structures. In some implementations, the model determines a first value for a first feature from a combination of values for one or more features before producing the output data representing the predicted entity of interest.

Claims Coverage

The provided independent claims cover three implementation types: a method for predicting an entity of interest from annotated pathology image feature values, a system performing the same prediction workflow using computer hardware processor instructions, and a non-transitory computer-readable medium storing instructions to perform the workflow. Across the independent claims, the inventive features include supervised extraction of feature values from annotated pathology images, prediction of patient clinical outcomes, and training using annotated training pathology images collected from patients participating in a randomized controlled clinical trial.

Overall, the independent claims consistently require extracting feature values from annotated pathology image input data using at least one machine learning model, predicting an entity of interest comprising patient clinical outcomes, outputting prediction output data, and training the model using extracted feature values from plurality of annotated training pathology images collected from patients participating in a randomized controlled clinical trial.

Stated Advantages

Documented Applications