Investigating the effect of masking and background field removal algorithms on the quality of QSM reconstructions using a realistic numerical head phantom.

Abstract

Background field removal (BFR) is an important step in the QSM pipeline, enabling the reconstruction of local susceptibility distributions by removing contributions from sources outside the region of interest (ROI). BFR requires calculation of a binary ROI mask, to which most BFR methods are sensitive. We investigated how masking, and errors in local field map estimation, impact the quality of QSM reconstructions. We used the 2019 QSM Reconstruction Challenge brain phantom to simulate multi-echo gradient echo acquisitions. Echoes were combined using complex fitting followed by unwrapping with SEGUE. Fifteen background field removal methods were applied using 4 local field masks. Local fields were compared with RMSE. Seven different QSM reconstruction algorithms were applied to the local fields and evaluated using the 2019 QSM Challenge metrics. For local field map estimation, PDF and MSMV performed best overall, although their performance was sensitive to the mask. V-SHARP and RESHARP were more robust to masking and showed good performance. LBV had low accuracy, which was improved by removing a polynomial fit. Surprisingly, this did not propagate to susceptibility, where LBV without polynomial fitting performed better. When paired with the Weak Harmonic QSM algorithm, LBV showed the best overall performance with low sensitivity to the mask; PDF and MSMV were next best. PDF and MSMV are robust choices for estimating local field maps and provide accurate QSM but can lead to susceptibility underestimation near brain boundaries. LBV is less reliable for local field map estimation but gives accurate results when used with weak harmonic QSM.