Deep Curriculum Learning for Follow-up MRI Registration in Glioblastoma

This paper presents a weakly supervised deep convolutional neural network-based approach to perform voxel-level 3D registration between subsequent follow-up MRI scans of the same patient. To handle the large deformation in the surrounding brain tissues due to the tumor's mass effect we proposed curriculum learning-based training for the network. Weak supervision helps the network to concentrate more focus on the tumor region and resection cavity through a saliency detection network. Qualitative and quantitative experimental results show the proposed registration network outperformed two popular state-of-the-art methods.