Code for the paper entitled "3D RoI-aware U-Net for Accurate and Efficient Colorectal Cancer Segmentation"(https://arxiv.org/abs/1806.10342).

The latest version of the 3D RU-Net code is implmented with PyTorch, which elegantly realizes the essential part of our algorithm and enables in-place computing.

Here are some results of colorectal cancer segmentation, which is the case of the paper; and illustrations of another task, mandible and masseter segmentation, showing the scalability of the proposed method.

Latest experiment: simultaneously segmenting 14 organs from pelvic CTs in ~0.5s (We trained this model with 24 training samples).

The code along with weights and a test fold are currently released.