Rapid Quantitative Susceptibility Mapping for Intracranial Hemorrhage using Deep Learning-based 2.5D Diffusion Models

Motivation: EPI-based Quantitative Susceptibility Mapping (QSM) benefits intracranial hemorrhage (ICH) imaging with reduced scan times for patients scan but faces limitations in image quality. Goal(s): a generalized deep learning QSM method that is robust to motion artifact, and capable of image quality boost from EPI acquisitions, while preserving the accuracy of ICH susceptibility quantification. Approach: A novel deep learning-based QSM method using 2.5D Diffusion Models (QSMDiff) is developed, with synthetic hemorrhagic susceptibility features extending the capability to ICH patients. Results: A deep learning QSM method robust to motion artifacts, enhancing EPI image quality while maintaining accurate ICH susceptibility quantification. Impact: This reliability across imaging conditions highlights QSMDiff's potential as a versatile and accurate tool for clinical susceptibility mapping for ICH patients.