A Framework to Generate Patient-Specific Numerical Models for Magnetic Resonance Imaging Simulations

Purpose: Anatomical models with comprehensive tissue properties are needed for realistic MRI simulation, yet no validated method exists to generate such models automatically from patient scans. This work presents a pipeline for generating patient-specific numerical brain models suitable for physics-based MRI simulation. Methods: A simulation-ready head model was constructed from the IT’IS Duke body model by harmonizing tissue labels through an anatomy ontology to assign MRI relaxation and proton density values from the IT’IS tissue property database. A multi-metric deformable registration framework was applied to deform the Duke model into subject-specific geometry for 30 healthy adults (17 male, mean age 45.3 years, range 24.3–68.1 years) with 1.5 T scans from the IXI dataset. Agreement between simulated and acquired images was evaluated using SSIM, normalized cross-correlation (NCC), and normalized root mean square error (NRMSE). Results: Average SSIM was 0.734 (p = 1.3 × 10−6), NCC was 0.747 (p = 3.8 × 10−3), and NRMSE was 0.116 (p = 9.4 × 10−9). Conclusion: The proposed framework requires no segmentation of the target image and automatically produces patient-specific models with comprehensive atlas-derived tissue properties suitable for synthetic MRI generation and coil simulation.