A Cross-Modality Attention Network for Automated Habenula Segmentation and Quantization by MP2RAGE in Patients With Major Depression Disorder

Motivation: The habenula is an important diencephalon structure related to pathophysiology of major depressive disorder (MDD), and segmenting Hb is challenging due to its low contrast and tiny size. Goal(s): To establish a 3D cross-modality attention network (CMAN) for Hb segmentation and quantify its volume and T1 value in MDD patients compared to healthy controls (HCs). Approach: We leveraged T1WI and T1 mapping from 3.0T MP2RAGE to train a CMAN model, assessing its accuracy and clinical relevance. Results: The CMAN model achieved high accuracy in segmenting Hb, revealing significant differences in T1 values between MDD patients and HCs. Impact: The habenula serves as a pivotal center modulating intricate behaviors encompassing emotions, motivation, and decision-making. Automatic and precise segmentation of habenula provides basis for further automatic quantitative analysis of habenula for major depression disorder studies.