HiFi-MambaV2: Hierarchical Shared-Routed MoE for High-Fidelity MRI Reconstruction

2025-11-23

AI Summary

To address high-frequency detail loss and anatomical distortion in undersampled k-space MRI reconstruction, this paper proposes a novel framework integrating frequency decomposition with content-adaptive computation. Methodologically, we design a hierarchical shared-routing Mamba-MoE architecture enabling pixel-wise sparse expert selection; introduce a separable frequency-domain consistent Laplacian pyramid; and combine unrolled network optimization, data consistency regularization, and lightweight global context fusion to jointly restore local textures and preserve global structural coherence. Evaluated on fastMRI and CC359 benchmarks, our approach significantly outperforms CNN-, Transformer-, and state-of-the-art Mamba-based methods, achieving substantial improvements in PSNR and SSIM, marked reduction in NMSE, and simultaneous enhancement of high-frequency detail fidelity and overall anatomical accuracy.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Computer Vision: Multi-modal Vision
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Reconstructing high-fidelity MR images from undersampled k-space data requires recovering high-frequency details while maintaining anatomical coherence. We present HiFi-MambaV2, a hierarchical shared-routed Mixture-of-Experts (MoE) Mamba architecture that couples frequency decomposition with content-adaptive computation. The model comprises two core components: (i) a separable frequency-consistent Laplacian pyramid (SF-Lap) that delivers alias-resistant, stable low- and high-frequency streams; and (ii) a hierarchical shared-routed MoE that performs per-pixel top-1 sparse dispatch to shared experts and local routers, enabling effective specialization with stable cross-depth behavior. A lightweight global context path is fused into an unrolled, data-consistency-regularized backbone to reinforce long-range reasoning and preserve anatomical coherence. Evaluated on fastMRI, CC359, ACDC, M4Raw, and Prostate158, HiFi-MambaV2 consistently outperforms CNN-, Transformer-, and prior Mamba-based baselines in PSNR, SSIM, and NMSE across single- and multi-coil settings and multiple acceleration factors, consistently surpassing consistent improvements in high-frequency detail and overall structural fidelity. These results demonstrate that HiFi-MambaV2 enables reliable and robust MRI reconstruction.

Problem

• Reconstructing high-fidelity MR images from undersampled k-space data
• Recovering high-frequency details while maintaining anatomical coherence
• Enabling reliable and robust MRI reconstruction across multiple settings

Innovation

• Hierarchical shared-routed MoE Mamba architecture
• Separable frequency-consistent Laplacian pyramid decomposition
• Lightweight global context path with data-consistency regularization