SHFormer: Dynamic Spectral Filtering Convolutional Neural Network and High-pass Kernel Generation Transformer for Adaptive MRI Reconstruction

2026-07-22

AI Summary

This work addresses the limitations of existing attention-based MRI reconstruction methods, which struggle to recover high-frequency details and exhibit poor generalization across modalities, often requiring separate training for each. To overcome these issues, the authors propose a multispectral attention mechanism that integrates a dynamic spectral filtering CNN with a high-pass kernel-generating Transformer. This architecture enables context-aware propagation of high-frequency information and cross-modal transferable feature extraction through neural modulation. By combining self-supervised and diffusion-based training strategies, the method mitigates the inherent low-frequency bias of conventional attention mechanisms and reduces modality dependence. Evaluated on unseen multimodal MRI data, the approach significantly improves reconstruction quality—yielding gains of approximately 1 dB in PSNR and 0.01 in SSIM—while demonstrating strong closed-set and open-set generalization capabilities.

Abstract

Attention Mechanism (AM) selectively focuses on essential information for imaging tasks and captures relationships between distant pixel neighborhoods to compute feature representations. Accelerated MRI reconstruction benefits from AM, as the imaging process involves Fourier domain measurements that influence image representation non-locally. However, AM-based models are more adept at capturing low-frequency information with limited capacity for high-frequency representations, restricting models to smooth reconstruction. Additionally, AM-based models need mode-specific retraining for multimodal MRI data, as their knowledge is restricted to local contextual variations that may be inadequate to capture transferable features across heterogeneous domains. To address these challenges, we propose a neuromodulation-based discriminative multi-spectral AM for scalable MRI reconstruction that can (i) propagate context-aware high-frequency details for high-quality reconstruction, and (ii) capture features reusable across deviated unseen domains in multimodal MRI. The proposed network consists of a spectral filtering CNN to capture mode-specific transferable features and a dynamic high-pass kernel generation transformer focusing on high-frequency details. We evaluate our model on comparative studies in supervised and self-supervised learning, diffusion model-based training, closed-set and open-set generalization under heterogeneous MRI data, and interpretation-based analysis. Our method offers scalable, high-quality reconstruction with best improvement margins of ~1 dB in PSNR and ~0.01 in SSIM under unseen scenarios. Code: https://github.com/sriprabhar/SHFormer

Problem

• MRI reconstruction
• attention mechanism
• high-frequency details
• multimodal MRI
• domain generalization

Innovation

• Dynamic Spectral Filtering
• High-pass Kernel Generation
• Cross-domain Generalization
• Multi-spectral Attention
• Adaptive MRI Reconstruction