Advancing Quantitative Susceptibility Mapping With 2.5D Diffusion Models for Rapid Intracranial Hemorrhage...

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Artificial Intelligence

Background:

• Quantitative susceptibility mapping (QSM) is crucial for assessing intracranial hemorrhage (ICH).
• Traditional QSM methods face challenges with rapid imaging techniques like echo planar imaging (EPI) and data scarcity.
• Developing robust and efficient QSM reconstruction algorithms is essential for clinical applications.

Purpose of the Study:

• To develop a generative diffusion model-based approach for robust and efficient quantitative susceptibility mapping (QSM) reconstruction in intracranial hemorrhage (ICH).
• To ensure applicability to both standard gradient echo (GRE) and rapid echo planar imaging (EPI) acquisitions.
• To address data scarcity challenges in QSM for ICH.

Main Methods:

• Proposed QSMDiff, an unsupervised diffusion model for 3D QSM dipole inversion.
• Evaluated volumetric partitioning strategies (2D slices, 3D patches, 2.5D slabs), adopting the memory-efficient 2.5D slab approach.
• Implemented a conditional sampling mechanism and a three-stage training data-generation strategy with synthetic ICH augmentation.

Main Results:

• QSMDiff demonstrated superior performance in simulations (SSIM: 0.97±0.07, RMSE: 0.04±0.03).
• Achieved strong agreement with SWI-QSM references in vivo for ICH patients (R²=0.83).
• Qualitative evaluation showed enhanced resolution and artifact suppression, especially in low SNR and motion conditions.

Conclusions:

• QSMDiff enables high-quality and accurate QSM reconstruction from both GRE and rapid EPI scans for ICH assessment.
• The 2.5D training strategy and synthetic data augmentation overcome limitations of lower-quality acquisitions.
• QSMDiff offers a practical solution for fast and robust ICH assessment.