A New Three-stage Curriculum Learning Approach for Deep Network Based Liver Tumor Segmentation

Abstract

Automatic segmentation of liver tumors in medical images is crucial for computer-aided diagnosis and therapy. It is a challenging task, since the tumors are notoriously small against the background voxels. This paper proposes a new three-stage curriculum learning approach for training deep networks to tackle this small object segmentation problem. The learning in the first stage is performed on the whole input volume to obtain an initial deep network for tumor segmentation. Then the second stage of learning focuses on the tumor-specific features by continuing training the network on the tumor patches. Finally, we retrain the network on the whole input volume in the third stage, in order that the tumor-specific features and the global context can be integrated to improve the final segmentation accuracy. With this approach, we can employ a single network to segment the tumors directly without the need of liver segmentation. We evaluate our approach on a clinical dataset from the hospital and the public MICCAI 2017 Liver Tumor Segmentation (LiTS) Challenge dataset. In the experiments, our approach exhibits significant improvement compared with the commonly used cascade counterpart.