Figure 3
From: A machine learning-based prognostic predictor for stage III colon cancer
A neural network (CNN) segmented and restored the H&E whole tissue slides. The CNN, InceptionResNetV2 was used to recognize the nine categories (ADI, adipose tissue; BACK, background; DEB, debris; LYM, lymphocyte; MUC, mucus; MUS, muscle; NORM, normal mucosa; STR, stroma; TUM, tumor) in each whole tissue slides from the Image Set B and C. Left panel showed the original H&E staining tissue slides, the right panel was the classification maps restored by CNN, the pie charts showed the proportions of each tissue category. (A) typical adenocarcinoma and (B) mucous adenocarcinoma were from the Image Set B. (C,D) were from the Image Set C. (C) showed some problems caused by handcraft, such as tissue fold and hollowing, (D) presented visualization problems caused by uneven fixation and covering of the slides. Despite these imperfections in the whole tissue slides, the trained CNN still can perfectly recognize the different tissue categories.
