Abstract 1629: Classification of cancer histology images using deep learning

Abstract

Abstract Pathological evaluation of tumor tissue images stained with hematoxylin and eosin (H&E) is pivotal and specialists often disagree on the final diagnosis. Meanwhile, automated image analysis approaches have great potential to increase precision of diagnosis, help reduce human error and cut cost. In this study, we utilize various computational methods based on deep learning framework and build a stand-alone diagnostic tool to effectively classify different histopathology images across different types. The dataset of tissue slides from eligible patients with breast cancer/ thyroid cancer consists of 45,000 samples in our hospital. Available tissue samples above were split into a training set, for learning the CNN parameters, and test set, for evaluating its performance. An accuracy of 94% was obtained for non-cancer (i. e. normal or benign) vs malignant (i. e. invasive carcinoma or papillary thyroid carcinoma). In addition, we also demonstrated the utility of our model to discriminate between two subtypes of lung cancer. It provides pathologists or medical technicians a straightforward platform to use without requiring sophisticated computational knowledge, and cancerization would be identified which is not visible under a single microscope. Citation Format: Weidong Xie. Classification of cancer histology images using deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1629.