Quantitative and Qualitative Impact of Magnetic Microstructure on Conventional and Deep Learning-based Quantitative Susceptibility Mapping

Motivation: Quantitative susceptibility mapping (QSM) is widely used to estimate iron and myelin contents from MRI phase data but is known to rely on simplistic assumptions about magnetic microstructure. Goal(s): To assess the impact of microstructure-induced phase shifts on susceptibility maps generated by various QSM algorithms, including recent deep learning methods. Approach: Using a realistic digital brain model and in vivo data, we compared conventional and deep learning-based QSM algorithms, assessing accuracy and anatomical consistency in susceptibility estimates. Results: Microstructure-induced phase shifts substantially affect the accuracy of susceptibility maps computed with different QSM techniques. Impact: This work demonstrates that incorporating microstructure-induced frequency shifts into QSM significantly enhances susceptibility mapping accuracy, particularly in white matter. These findings enable improved study of pathologies with microstructural alterations, providing clinicians and researchers with more reliable maps for disease assessment.