Radiogenomics for cancer subtype feature visualization

Abstract

A computer implemented method and system for identifying and visualizing features correlated with cancer subtypes, includes the steps of receiving a set of medical images representing an area including a patient's lesion, applying a classification model trained on lesions with known pathology by mathematically combining the model's learned weights with said medical images to produce a score indicating the likelihood of the lesion being of each particular cancer subtype, extracting importance maps from said classification model indicating features correlated with each cancer subtype by tracing the output of the model through the learned weights to identify regions of the input with a strong influence on the classification score for each cancer subtype, and displaying said importance maps visually on a computing device.

Dependencies are inferred from the claim text.