Segmentation and volume estimation of habenula using deep-learning in patients with depression

Summary

　The research group utilized deep learning with manually segmented images as training data to develop a model capable of performing rapid and automatic segmentation of the habenula, a putative key structure of depression, on 3 Tesla MR images. Following validation of the model using more precise 7 Tesla MR images of the same patients, the correlation between habenula volume and clinical variables was investigated, with the dataset obtained in the Strategic International Brain Science Research Promotion Program (Brain/MINDS Beyond) (234 patients with depression, 148 healthy individuals). The habenula volume exhibited a negative correlation with age across all groups, which was more pronounced in the depression group. Additionally, after controlling for age and inter-scanner differences, an inverse correlation of habenula volumes with severity of depression was demonstrated in female patients. These findings suggest that habenula volume may serve as a diagnostic marker of depression, particularly in females.

Article