VoxStruct3D: Structure-Leading Flow Matching for Voxel-Space 3D MRI Synthesis

2026-08-05

AI Summary

Existing high-fidelity 3D MRI synthesis methods are constrained by the reconstruction bottlenecks of autoencoders, struggling to simultaneously preserve global anatomical consistency and voxel-level detail. This work proposes VoxStruct3D, a framework that directly models full-resolution MRI volumes in voxel space through an innovative “structure-first, image-follows” mechanism. Structural token trajectories guide image generation, while Patch-Aligned RoPE and asymmetric attention enforce unidirectional structural constraints on the synthesized images. The model employs a flow-matching objective and integrates a voxel generator—featuring factorized 3D patch embeddings, overlapping upsampling, time-modulated residual refinement, and skip fusion—with a StructVAE and a pretrained 3D medical encoder. Evaluated on both pathological and healthy T1-weighted brain MRI data, the method significantly outperforms existing approaches in feature distribution alignment, sample diversity, and perceptual quality.

Abstract

High-fidelity 3D MRI synthesis requires both globally coherent anatomy and fine-grained voxel-level detail. Although latent diffusion makes volumetric generation tractable, its image autoencoder introduces a reconstruction bottleneck that can limit the fine detail recoverable in the final volume. We present VoxStruct3D, a voxel-space flow-matching framework that directly models full-resolution MRI volumes using a clean-data prediction objective. Its Volumetric Voxel Generator (VVG) combines factorized 3D patch embedding with overlapping upsampling, time-modulated residual refinement, and skip fusion, enabling neighboring tokens to jointly reconstruct shared voxel regions and suppress patch-boundary artifacts. To complement direct voxel-space modeling with an explicit anatomical prior, we further introduce a Structure-First, Image-Follows (SFIF) strategy. A frozen pretrained 3D medical encoder and a StructVAE extract compact structure tokens that preserve dominant anatomy, while a structure-leading schedule keeps their trajectory ahead of the image trajectory. Patch-Aligned RoPE spatially aligns the unequal token grids, and asymmetric attention enforces one-way guidance from structure to image. Experiments on pathological and healthy T1-weighted brain MRI datasets show that VoxStruct3D achieves the strongest overall performance across feature-distribution alignment, sample diversity, and perceptual quality, producing anatomically coherent and visually realistic volumes.

Problem

• 3D MRI synthesis
• voxel-level detail
• anatomical coherence
• reconstruction bottleneck
• high-fidelity generation

Innovation

• voxel-space generation
• flow matching
• structure-leading guidance
• 3D MRI synthesis
• patch-aligned RoPE