Magnetic susceptibility source separation technology: a narrative review of principles, validation, and clinical translation

Quantitative susceptibility mapping (QSM) returns the net susceptibility of each voxel. When paramagnetic and diamagnetic materials coexist, their opposing contributions can cancel, and QSM cannot recover them separately. Magnetic susceptibility source separation (MSSS) adds a relaxation-sensitive measurement, usually R2′ or a calibrated estimate derived from R2*, to the signed local field. The combined data are used to estimate paramagnetic- and diamagnetic-associated components. In this narrative review, we examine the acquisition demands, physical assumptions, validation, and clinical use of the main MSSS methods. We consider R2′-based χ-separation, the GRE-only DECOMPOSE-QSM and R2*QSM methods, adaptive-parameter techniques, and deep-learning pipelines. Simulations and phantom studies support technical feasibility, and a smaller body of ex vivo and in vivo work provides biological and clinical evidence. These maps nevertheless remain model-dependent surrogates, not direct measurements of iron or myelin concentration. Interpretation is affected by susceptibility anisotropy, source geometry, diffusion, multicompartment water, errors in R2 estimation, and differences between scanners and protocols. Early longitudinal studies in multiple sclerosis (MS) have linked component changes to lesion repair. Evidence on paramagnetic rim lesions, however, still comes mainly from conventional phase or QSM imaging, and MSSS-specific clinical data are limited. Studies in Parkinson’s and Alzheimer’s disease are exploratory. Before routine clinical use, MSSS needs real-world validation, anisotropy-aware modeling, harmonized multicenter acquisition, and independent longitudinal testing.