Comparing Voxel- and Streamline-Based Paradigms In Clinical Tractography
Abstract
Purpose
Tractography has become more widely used in neurosurgical planning, yet traditional voxel-based Region of Interest (ROI) segmentation often includes anatomically implausible fibers and becomes challenging with distorted anatomy near lesions. This study evaluates the paradigm shift from voxel- to streamline-based tractography for reconstructing the Arcuate Fasciculus (AF) and Inferior Fronto-Occipital Fasciculus (IFOF), using Constrained Spherical Deconvolution (CSD) with both deterministic and probabilistic tracking. Both approaches are evaluated for robust, clinically meaningful bundle reconstructions.
Methods
Tractograms were generated using deterministic and probabilistic CSD methods, with 0.5 to 8M streamlines, from ten patients diffusion MRI data. Bundles were segmented using BundleSeg with the TractSeg (voxel-wise) and Yeh (streamline-based) atlases. Morphological fidelity was assessed using volume, shape descriptors (elongation, diameter, fractal dimension, irregularity), and Bundle Shape Similarity (BSS). Comparisons were performed across atlas paradigms, tractography methods, and streamline numbers.
Results
Yeh atlas produced compact, coherent bundles, while TractSeg atlas generated larger volumes with less anatomically constrained fibers. Trunk volumes are approximately stable across streamlines number except for probabilistic voxel-wise atlas. Shape metrics confirm the intrinsic difference of the two atlases, with lower surface irregularity and fractional dimension for the streamline-based approach, emphasized by higher average length and elongation, then enhanced continuity. Deterministic CSD achieved higher shape fidelity with clinical references (BSS ≈0.6–0.8) compared to probabilistic methods (BSS ≈0.2–0.4). Shape and volume metrics stabilized ~2M streamlines. Voxel- and streamline-based segmentations are complementary: volume for safety, streamlines for tract geometry.
Conclusion
Depending on clinical context, deterministic CSD method with ~2M streamlines and well-defined atlas (Yeh or TractSeg) provides the most reliable and interpretable configuration. Using both voxel- and streamline-based atlases allows case-by-case selection: tight reconstructions for limited space, voluminous for small lesions’ safety. A quantitative, physics-based approach to developing clinically validated tractography reconstructions improves surgical safety and preserves critical neural connections.