Self-Supervised Spatial And Zero-Shot Angular Super-Resolution by Spatial-Angular Implicit Representation For Rotating-View SNR-Efficient Diffusion MRI

2026-05-04

AI Summary

This work addresses the challenge of long acquisition times in thick-slice diffusion MRI (dMRI) with high angular sampling, despite its high signal-to-noise ratio. The authors propose a spatial–angular implicit neural representation (SA-INR) that leverages a b=0 structural prior and diffusion-encoding directions as conditioning inputs to a FiLM-modulated MLP, enabling end-to-end reconstruction of high-resolution dMRI from only a single observation per diffusion direction. The method supports zero-shot synthesis for unseen diffusion directions and jointly achieves spatial super-resolution and angular zero-shot super-resolution—surpassing the Nyquist sampling limit for the first time. Experiments demonstrate reconstruction PSNRs of 34.82 dB and 33.08 dB for seen and unseen directions, respectively, on simulated data, while significantly improving quantitative accuracy in downstream diffusion tensor imaging (DTI) models.

Abstract

Rotating-view thick-slice acquisition is highly SNR-efficient for mesoscale diffusion MRI (dMRI) but requires numerous rotating views to satisfy Nyquist sampling, resulting in long scan time. We propose a self-supervised Spatial-Angular Implicit Neural Representation (SA-INR) that reconstructs high-resolution dMRI from a single view per diffusion direction, representing a massive acceleration. Our model, an MLP conditioned on a b=0 structural prior and the b-direction via FiLM, is trained end-to-end on the anisotropic input. The framework not only accurately reconstructs the trained b-directions (spatial SR) but also learns a continuous q-space representation, enabling high-fidelity "zero-shot" synthesis of unseen b-directions (angular SR). On simulated data, our method achieved high fidelity for both trained (34.82 dB) and unseen (33.08 dB) directions. Most importantly, the synthesized angular data also improved the quantitative accuracy of downstream DTI model fitting. Our SA-INR framework breaks the classical sampling limits, paving the way for fast, quantitative high-resolution dMRI.

Problem

• diffusion MRI
• super-resolution
• rotating-view acquisition
• SNR efficiency
• Nyquist sampling

Innovation

• Self-supervised learning
• Implicit Neural Representation
• Zero-shot angular super-resolution
• Diffusion MRI
• SNR-efficient acquisition