Brain Child

27 April 2018

Automatic Registration and Segmentation of 3D Brain Images

Area: Deep Learning; Technologies: PyTorch, GCP.

The project examines the efficacy of different deep learning techniques for registering and segmenting 3-D brain magnetic resonance images (MRI) from 100 different subjects. The goal is to develop separate networks that can be used to complete the tasks of registration and segmentation. This will ultimately help researchers better quantify differences between health and pathogenic brains as well as better understand early disease markers in brain disorders. To this end, our research implements two different networks, a version of fully convolutional neural networks for registration, and the V-net architecture for segmentation. While we were not able to train the V-net architecture due to memory limitations, we were able to successfully train 1000 epochs of the registration network. The project was accomplished in a team of 2.