AI-driven segmentation of the anterior visual pathway from high-resolution MR images: development and clinical validation in MS patients.

Motivation: Anterior Visual Pathway (aVP) abnormalities are linked to various etiologies, however the current lack of a standardized and automated framework renders analysis laborious and inconsistent. Goal(s): To develop an AI-driven tool for automated aVP segmentation, and clinically validate in healthy controls (HC) and multiple sclerosis (MS) patients. Approach: 0.6 mm isotropic 3D constructive interference steady-state images from 40 HC and 49 MS patients were segmented using a high-resolution 3D V-Net model. Results: Segmentations' spatial similarity between AI and both neuroradiologists was good (average DSC 0.8, CI 95% CI 0.77 - 0.83). Both AI and neuroradiologists' results could discriminate between HC and MS. Impact: This study demonstrates the clinical potential of AI-driven segmentation, enhancing the efficiency and preserving the accuracy, of MRI-based structural integrity of the anterior visual pathway in multiple sclerosis, thus paving the way for more consistent and reliable diagnostic workflows.