MINeR: Direction-modulated implicit neural representation enables ultrafast multi-shell diffusion MRI.
Authors
Affiliations (4)
Affiliations (4)
- School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, 200030, China.
- National Key Laboratory of Advanced Micro and Nano Manufacture Technology, Shanghai Jiao Tong University, Shanghai, 200240, China.
- School of Information Science and Technology, ShanghaiTech University, Shanghai, 201210, China. Electronic address: [email protected].
- School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, 200030, China; National Engineering Research Center of Advanced Magnetic Resonance Technologies for Diagnosis and Therapy (NERC-AMRT), Shanghai Jiao Tong University, Shanghai, 200030, China. Electronic address: [email protected].
Abstract
Diffusion magnetic resonance imaging (dMRI) enables noninvasive mapping of tissue microstructure by probing water molecule diffusivity. While advanced multi-shell diffusion models offer improved sensitivity to cellular properties, their requirement for densely sampled q-space data leads to prohibitively long acquisition times. Current deep learning approaches for parameter estimation face three key limitations: (1) dependency on fixed acquisition protocols, (2) model-specific assumptions that constrain applicability, and (3) reliance on supervised learning paradigms that demand large labeled datasets and exhibit poor generalization to out-of-distribution cases. To address these challenges, we propose MINeR, a novel unsupervised subject-specific framework for reconstructing dense q-space data from highly undersampled acquisitions. Our method leverages direction-modulated implicit neural representation to flexibly sample diffusion signals across q-space, supporting the estimation of parameters for diverse diffusion models. Comprehensive evaluations demonstrate that MINeR maintains high fidelity in microstructural parameter estimation, particularly for advanced multi-shell diffusion models. The framework shows remarkable generalization capability, as evidenced by its robust performance on tumor data. Notably, MINeR effectively reconstructs high-quality diffusion signals by interpolating from 6 directions, significantly reducing acquisition time, while maintaining robust parameter estimation. This work presents a practical approach for enabling microstructural modeling from sparsely sampled q-space data, thereby improving the clinical applicability of diffusion MRI. The code is available at: https://github.com/AMRI-Lab/MINeR.