Systems and methods for assessing liver pathology

Inventors

Taylor-Weiner, Amaro N. • Pokkalla, Harsha Vardhan • Elliott, Hunter L. • Glass, Benjamin P. • Wapinski, Ilan N. • Khosla, Aditya • Resnick, Murray • Montalto, Michael C. • Beck, Andrew H. • Shanis, Zahil • Pedawi, Aryan • Le, Quang Huy • Wang, Jason K. • POURYAHYA, Maryam • Leidal, Kenneth Knute • Carrasco-Zevallos, Oscar M. • Juyal, Dinkar • Biddle-Snead, Charles • WACK, Katy

Assignees

PathAI Inc

Abstract

In some aspects, the described systems and methods provide for a method for training a deep learning model to assess liver pathology, including accessing annotated liver pathology images associated with a group of patients in one or more randomized controlled clinical trials of nonalcoholic steatohepatitis therapy, each of the annotated liver pathology images including at least one annotation describing one or more tissue characteristic categories for a portion of the image, and training the deep learning model based on the annotated liver pathology images to predict the tissue characteristic categories, selected from a group comprising steatosis, lobular inflammation, hepatocyte ballooning, and fibrosis stage.

Core Innovation

The invention relates to systems for training a deep learning model to assess liver pathology using annotated liver pathology images associated with a group of patients in one or more randomized controlled clinical trials of nonalcoholic steatohepatitis therapy. The annotated images include annotations describing one or more tissue characteristic categories for a portion of the image, and the system trains the deep learning model to predict those tissue characteristic categories for a liver pathology image. The tissue characteristic categories are selected from steatosis, lobular inflammation, hepatocyte ballooning, and fibrosis stage, including a correspondence where steatosis, lobular inflammation, and hepatocyte ballooning correspond to components of a nonalcoholic fatty liver disease activity score (NAS).

For evaluation, the system compares a fraction of tissue area in the liver pathology image assigned to each NAS component with an ordinal score determined by a pathologist based on the NAS components. The invention stores the trained deep learning model on at least one storage device for subsequent use in evaluating performance on a liver pathology image, and states that intra-observer reproducibility of the ordinal score by multiple pathologists is lower than reproducibility of predictions from multiple iterations of the trained deep learning model processing the liver pathology image.

Additional aspects refine the approach by using deep convolutional neural networks to form the deep learning model and by generating slide-level continuous fibrosis scoring that reflects underlying heterogeneous patterns of fibrosis from predicted tissue characteristic categories. Further refinements include pixel-level prediction of tissue characteristic categories, extraction and use of complex feature values from annotated liver pathology images, and determining a response-metric score by computing a distance between distributions of fibrosis stages at baseline and following therapy, with optional classification of responder and non-responder groups.

Claims Coverage

The partial content identifies one independent claim (clm-00001). It covers a computer-based system that trains a deep learning model on annotated liver pathology images from randomized controlled clinical trials of nonalcoholic steatohepatitis therapy, predicts tissue characteristic categories, and evaluates performance by comparing predicted tissue-area fractions assigned to NAS components with an ordinal pathologist-derived NAS score, including a stated reproducibility relationship. Dependent claims (not fully provided) further specify refinements such as deep convolutional neural networks, slide-level continuous scoring, pixel-level prediction, complex feature extraction, and a distance-based response metric.

Across the provided partial content, the independent claim centers on a system that trains a deep learning model from annotated liver pathology images tied to NASH randomized controlled clinical trials, predicts NAS-related tissue characteristic categories and fibrosis stage, and evaluates performance by tissue-area fractions versus an ordinal pathologist NAS score, with a stated reproducibility relationship. The associated refinements described in the partial content include model architecture as deep convolutional neural networks, slide-level continuous fibrosis scoring, pixel-level prediction, complex feature values, and a distance-based fibrosis-stage response metric.

Stated Advantages

Documented Applications