Region-of-Interest Based Statistical Analysis of the 2019 QSM Challenge

Since its inception, the most recent QSM challenge and corresponding dataset have helped to improve dipole-inversion algorithms. To date, only global metrics have been used to evaluate performance of these algorithms. Typically, in clinical applications of QSM, the susceptibility value in a specific brain region or structure is of interest. Therefore, we compared the accuracy of QSM algorithms in 28 regions in several ways including regional metrics and median differences, which is important to facilitate clinical translation of QSM. We found that, according to regional metrics, some direct QSM methods are surprisingly accurate considering their low global scores.