Magnetic susceptibility source separation (χ-separation) in quantitative susceptibility mapping

Conventional quantitative susceptibility mapping (QSM) outputs a voxel-averaged value, frequently obscuring co-localized paramagnetic iron and diamagnetic myelin through phase cancellation. χ-separation resolves this ambiguity by disentangling sub-voxel magnetic sources to quantify specific biomarkers. This review categorizes separation methodologies into multi-sequence approaches utilizing R2 or R2' (e.g., SEMI-TWInS, APART-QSM) and clinically expedient gradient-recalled echo (GRE)-only techniques (e.g., DECOMPOSE-QSM, QSM-ARCS). We further evaluate deep learning frameworks, such as χ-sepnet, designed to synthesize missing relaxation parameters for direct source mapping. Analysis reveals a trajectory toward simplified acquisition protocols, yet accurate determination of relaxometric constants and adherence to static dephasing regime assumptions remain significant biophysical hurdles. While specificity for iron and myelin quantification is improved, the absence of a definitive in vivo gold standard complicates absolute accuracy assessment. Future implementation must prioritize anisotropy-aware modeling and robust ground-truth validation to ensure clinical viability.