Speaker
Description
Tractography enables the reconstruction of white matter pathways from diffusion MRI and is a key tool for studying brain connectivity in both research and clinical contexts. Within the overall tractography pipeline, the parcellation step assigns individual streamlines to specific anatomical bundles, or discards them as false positive detections. We introduce PETParc (Parallel Efficient Tractography Parcellation), the first transformer-based architecture for this task. We demonstrate that treating individual streamlines as tokens, and letting them exchange information via the self-attention mechanism provides state-of-the-art results for registration-free parcellation, while being two orders of magnitude faster than TractCloud, the previous state of the art. Operating directly in subject space, PETParc avoids costly registration, generalizes well to unseen healthy subjects, and even outperforms TractCloud in pathological cases. In particular, we test robustness on subjects post-hemispherotomy, where PETParc reconstructs more complete and anatomically plausible bundles. Therefore, our approach offers a scalable and robust solution for high-throughput tractography analysis.
