Physics-Guided Diffusion Transformer with Spherical Harmonic Posterior Sampling for High-Fidelity Angular Super-Resolution in Diffusion MRI

2025-09-07

AI Summary

Diffusion MRI (dMRI) angular super-resolution (ASR) aims to reconstruct high-fidelity high-angular-resolution (HAR) signals from low-angular-resolution (LAR) acquisitions, yet existing methods suffer from coarse q-space geometric modeling and insufficient integration of biophysical constraints. This paper proposes a physics-guided diffusion Transformer framework that jointly incorporates a q-space geometry-aware module and a two-stage spherical harmonic posterior sampling scheme, enabling orientation-aware representation learning and physically interpretable signal reconstruction. To ensure training stability and physical consistency, we introduce b-vector modulation, random angular masking, and thermal-diffusion regularization. Evaluated on multiple public dMRI datasets, our method achieves superior fine-grained angular detail recovery, effectively suppressing over-smoothing and reconstruction artifacts. It outperforms state-of-the-art approaches across ASR, diffusion tensor imaging (DTI), and neurite orientation dispersion and density imaging (NODDI) tasks, providing a more reliable tool for microstructural characterization in neuroscience and clinical neuroimaging.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Multimodal Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Diffusion MRI (dMRI) angular super-resolution (ASR) aims to reconstruct high-angular-resolution (HAR) signals from limited low-angular-resolution (LAR) data without prolonging scan time. However, existing methods are limited in recovering fine-grained angular details or preserving high fidelity due to inadequate modeling of q-space geometry and insufficient incorporation of physical constraints. In this paper, we introduce a Physics-Guided Diffusion Transformer (PGDiT) designed to explore physical priors throughout both training and inference stages. During training, a Q-space Geometry-Aware Module (QGAM) with b-vector modulation and random angular masking facilitates direction-aware representation learning, enabling the network to generate directionally consistent reconstructions with fine angular details from sparse and noisy data. In inference, a two-stage Spherical Harmonics-Guided Posterior Sampling (SHPS) enforces alignment with the acquired data, followed by heat-diffusion-based SH regularization to ensure physically plausible reconstructions. This coarse-to-fine refinement strategy mitigates oversmoothing and artifacts commonly observed in purely data-driven or generative models. Extensive experiments on general ASR tasks and two downstream applications, Diffusion Tensor Imaging (DTI) and Neurite Orientation Dispersion and Density Imaging (NODDI), demonstrate that PGDiT outperforms existing deep learning models in detail recovery and data fidelity. Our approach presents a novel generative ASR framework that offers high-fidelity HAR dMRI reconstructions, with potential applications in neuroscience and clinical research.

Problem

• Reconstructing high-angular-resolution MRI signals from limited low-resolution data
• Overcoming inadequate q-space geometry modeling and physical constraint limitations
• Mitigating oversmoothing and artifacts in diffusion MRI angular super-resolution

Innovation

• Physics-guided diffusion transformer with q-space geometry awareness
• Spherical harmonics posterior sampling for data alignment
• Heat-diffusion regularization for physically plausible reconstructions