DeepPATH

The study trained an Inception v3 deep convolutional neural network on whole-slide images from The Cancer Genome Atlas to automatically classify lung tissue as adenocarcinoma (LUAD), squamous cell carcinoma (LUSC), or normal. The model achieved pathologist-comparable performance with an average AUC of 0.97 and was validated on independent datasets including frozen, FFPE, and biopsy samples. It was also able to predict several common LUAD gene mutations directly from histology images, indicating potential for assisting in both subtype classification and molecular inference.