Fig. 1

Overview of the proposed method. It aims to extract prognostic features from whole slide images and predicts patients’ survival risk. Our approach consists of four main parts: A sampling patches from whole slide images; B use of DeepConvSurv models to extract histopathological features of the tumor, lymphocyte, stroma, and mucus; C extraction of tissue area features from the tissue map by considering the tumor, lymphocyte, and stroma area; D training of several survival models using extracted features to predict patients’ risk. LYM lymphocyte, MUC mucus, STR stroma, TUM tumor, SSVM survival support vector machine, RSF random survival forest, GBRT gradient boosted regression tree