Comparing Repeatability Metrics for Quantitative Susceptibility Mapping in the Head and Neck

Motivation: To enable quantitative susceptibility mapping (QSM) in the head and neck, it is necessary to optimize a reconstruction pipeline that is repeatable across acquisitions. Metrics used to quantify QSM repeatability vary significantly and can produce contradictory results. Goal(s): To compare commonly used metrics of repeatability in a repeated acquisitions QSM dataset. Approach: We calculated root mean squared error, XSIM, standard deviation, repeatability coefficient (RC), and intraclass correlation coefficient (ICC) for QSM reconstructions using six state-of-the-art dipole inversion techniques. Results: Joint consideration of RC, ICC, and XSIM offered the best metric of repeatability for comparisons between QSM methods. Impact: This study shows that the repeatability of QSM in the head and neck depends strongly on the dipole inversion method used. Metrics like repeatability coefficient and intraclass correlation coefficient offer the most utility for comparisons between methods.