Simulation of realistic brain phantoms for susceptibility tensor imaging

To propose three Susceptibility Tensor Imaging (STI) brain phantoms as ground truth for evaluating STI reconstruction algorithms: two derived from STI data and the one from Diffusion Tensor Imaging (DTI).

The eigenvalues were generated through a pipeline inspired by the Quantitative Susceptibility Mapping (QSM) Reconstruction Challenge 2.0. An eigendecomposition was applied to an acquired susceptibility tensor, and literature-reported mean eigenvalues were assigned to 13 distinct brain regions. Fractional Anisotropy (FA) maps provided realistic spatial texture in the eigenvalues. Eigenvectors were obtained from DTI and two STI reconstructions: a Least-Squares algorithm and Diffusion Regularized STI (DRSTI). Microstructural simulations were incorporated into white matter regions. Phantom behavior was tested in three experiments: (i) anisotropic susceptibility assessment using QSM at different orientations; (ii) STI reconstructions across five algorithms with varying orientation numbers; and (iii) reconstructions under different angular rotation ranges.

The phantoms showed strong contrast in subcortical regions and realistic microstructural patterns. Algorithmic performance trends matched prior reports, with reduced errors as orientation count and rotation range increased.

Three open-source, in-silico STI brain phantoms were developed and validated as ground truth references for the evaluation of current and future STI reconstruction algorithms.