Two Parallel Stages Deep Learning Network for Anterior Visual Pathway Segmentation

The segmentation of the anterior visual pathway(AVP) from MRI sequences is challenging because of the thin long architecture, structural variations along the path, and poor contrast with adjacent anatomic structures. The AVP plays a critical role in many devastating pathological conditions (e.g., pituitary tumors and craniopharyngiomas). However, most of the existing methods segment AVP on T1w images merely and often fail to achieve good results that cannot meet clinical needs. In this work, we introduced fractional anisotropy(FA) images into the training data set and proposed a deep learning network with two parallel stages for AVP segmentation. On an MRI dataset consisting of 102 subjects selected from the Human Connectome Project (HCP), we demonstrate that the proposed framework consistently improves the accuracy of AVP segmentation.