Bi-parametric Joint Label Fusion: An Improved Segmentation Tool for Deep Gray Matter in QSM

Motivation: Publicly available T1-weighted-based segmentation tools may introduce systematic errors, especially in conditions like multiple sclerosis (MS), where atrophy is common. Goal(s): We aimed to improve segmentation accuracy by developing a multi-atlas, bi-parametric DGM tool (bi-parametric joint label fusion [B-JLF], leveraging QSM and T1-weighted images. Approach: We generated nine group-specific atlases across age and disease spectra, and modified the original ANTs JLF for multi-contrast based segmentation. We compared our tool against T1-based methods using overlap and susceptibility measures against manual segmentations. Results: B-JLF achieved superior voxel overlap (75%) and highest correlation with ground truth susceptibility (R²=0.88), outperforming T1-based methods. Impact: B-JLF provides reliable DGM segmentation, enhancing quantitative accuracy in neurodegenerative studies. The tool and atlases are publicly accessible, supporting broader neuroimaging applications.