An FA-based hybrid DTI-mixture model for multi-fiber reconstruction in diffusion MRI

This study proposes a hybrid diffusion MRI reconstruction framework combining Diffusion Tensor Imaging (DTI) and the multi-compartment non-central Wishart (MNCW) mixture model. The proposed framework uses an empirically selected fractional anisotropy (FA) threshold of 0.75 as a heuristic to distinguish between single- and multi-fiber voxels, enabling adaptive selection of the most appropriate reconstruction model for each voxel.

Synthetic diffusion MRI datasets containing single-, two-, and three-fiber configurations were generated using 82 diffusion-sensitizing gradient directions with Rician noise. Voxels with FA $\ge$ 0.75 were reconstructed using DTI, while voxels with FA < 0.75 were analyzed using the MNCW model. Angular error and clustering behavior were evaluated under different thresholds. The method was further validated using rat optic chiasm and human brain datasets.

The proposed framework reduced incorrect single-fiber detections and produced fewer disoriented voxels in complex fiber configurations compared with the conventional mixture model. A clustering threshold of 17 $_{}^{\circ }$ provided stable reconstruction without merging distinct fibers. Real data experiments demonstrated anatomically consistent reconstruction of white matter structures.

The hybrid framework improves orientation estimation by selectively applying mixture modeling only in voxels likely to contain multiple fibers, reducing over-segmentation, improving stability, and lowering computational complexity.