Repeatability of susceptibility separation methods in brain at 3 T

Motivation: Susceptibility separation methods aim to separate co-existing myelin and iron contributions. However, their repeatability has not been investigated. Goal(s): Evaluating repeatability of existing separation methods in brain and comparing their performance. Approach: Three methods (χ-Sep, χ-SepNet, and APART) were applied to 3T scan-rescan data of 21 healthy subjects, and the resultant dia- and paramagnetic maps were evaluated in white and deep gray matter regions. Results: Reliability varied by method and region, with many regions showing moderate to good reliability. Average repeatability coefficients were 4 ppb and 8 ppb in white matter and iron-rich deep gray regions, respectively. Impact: Susceptibility separation methods showed moderate to good reliability in most brain regions, and sub-voxel changes around 5 ppb might be error. Comparing values reported using different methods might not be straightforward, as the difference between measurements could exceed 15 ppb.