Direction-Based Latent Implicit Neural Representation for Diffusion MRI Parameters Mapping

Motivation: Advanced diffusion models for mapping diffusion parameters demand dense q-space sampling, increasing scan times. Current methods often require large training datasets, show limited generalization, and are constrained by fixed directions and b-values. Goal(s): To enable precise DWI interpolation across multiple directions and b-values in q-space for high-quality diffusion parameters mapping with sparse sampling. Approach: We propose an unsupervised framework using a direction-based Latent INR with adaptive weight modulation, allowing flexible q-space interpolation and robust parameters mapping with fewer directions. Results: Our model provided reliable parameters estimation and outperformed the traditional method, showing potential for efficient dMRI analysis across various diffusion models. Impact: This study enables efficient and adaptable diffusion parameter mapping, potentially reducing scan times and improving diagnostic workflows. The framework's flexibility may inspire new research on dMRI applications and help scientists explore a wider range of diffusion models with limited data.