Gastric Pathology Image Classification Using Stepwise Fine-Tuning for Deep Neural Networks

Figure 1

The proposed stepwise fine-tuning architecture. Apart from “low-level” pretraining datasets and well-annotated “high-level” datasets, “medium-level” data including tissue-wise data and cell-wise data are involved as well. In CNN models corresponding to training step t = 0, t = 1, and t = 2, blue nodes coloured darker denote more specified (deeper) representation for the pathology image classification task.