Whole Slide Image Classification of Gastric Cancer using Convolutional Neural Networks

Junni Shou, Yan Li, Guanzhen Yu, Guannan Li

Abstract: Gastric cancer is one of the main causes of cancer and cancer-related mortality worldwide, and the diagnosis based on histopathology images is a gold standard for gastric cancer detection. However, manual diagnosis is labor-intensive and low in inter-observer agreement. Computer-aided image analysis method were thus developed to alleviate the workload of pathologists and overcome the problem of subjectivity. Histopathology image analysis using deep learning has been proved to give more promising results than traditional methods on many whole slide image cancer detection tasks, including breast cancer detection and prostate cancer detection. In this paper, we further studied a whole slide image classification method using Convolutional Neural Networks (CNNs) on gastric cancer data. The method classify a whole slide image based on patch-sized classification results. Various experiments for patch-level classification using different existing CNN architectures were conducted. Experiment results show that the architecture gives the state-of-the-art result in natural image classification tasks can also give impressive results in histopathology image classification tasks.