Anatomy-to-Tract Mapping: Inferring White Matter Pathways Without Diffusion Streamline Propagation
Diffusion tractography, a cornerstone of white matter mapping, relies on point-to-point stream-line propagation--a process often compromised by errors stemming from inadequate signal-to-noise ratio and limited spatioangular resolution in diffusion MRI (dMRI) data. Here, we introduce Anatomy-to-Tract Mapping (ATM), the first model to our knowledge that generates bundle-specific streamlines directly from T1-weighted MRI without requiring orientation field estimation, voxelwise segmentation, and streamline propagation. ATM leverages the superior quality and minimal distortion of anatomical MRI and learns from multi-subject datasets to deliver robust, subject-specific streamline bundles with accurate preservation of structural connectivity. Trained on paired T1w and tractogram data, ATM learns to synthesize anatomically plausible streamlines conditioned on subject anatomy. This paradigm-shifting approach overcomes challenges associated with complex configurations, such as crossing, kissing, bending, and bottlenecks, providing anatomically guided bundle reconstructions. Using the TractoInferno dataset with 30 white matter bundles, we compared the performance of ATM against methods based on diffusion MRI, including MRtrix probabilistic tracking with BundleSeg for bundle segmentation, and Sherbrooke Connectivity Imaging Lab (SCIL) white matter atlas warping. ATM consistently showed strong performance across several metrics, including bundle similarity, volume coverage, angular correlation, streamline validity, geometric fidelity, and connection topology. ATM complements diffusion tractography by leveraging global anatomical features that are less susceptible to local uncertainties, providing a robust, anatomy-driven approach to reconstructing white matter pathways.