Mamba-Based Modality Disentanglement Network for Multi-Contrast MRI Reconstruction

2025-12-22

AI Summary

Existing accelerated MRI methods suffer from two key limitations: insufficient exploitation of k-space priors, leading to persistent aliasing artifacts, and interference from irrelevant modalities during multi-contrast fusion, degrading reconstruction quality. To address these, we propose a dual-domain cooperative modality-decoupled reconstruction framework. First, target k-space is completed using fully sampled reference data; then, a state-space model (Mamba) is integrated to implement modality-decoupled attention and iterative feature purification in the image domain, under structural alignment constraints that suppress modality-specific interference. This work pioneers the application of Mamba to multi-contrast MRI reconstruction, enabling joint k-space and image-domain optimization. Evaluated on multiple public datasets, our method achieves PSNR gains exceeding 2.1 dB over state-of-the-art approaches, significantly suppresses mixed artifacts, and demonstrates clinical diagnostic consistency validated by radiologists.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Magnetic resonance imaging (MRI) is a cornerstone of modern clinical diagnosis, offering unparalleled soft-tissue contrast without ionizing radiation. However, prolonged scan times remain a major barrier to patient throughput and comfort. Existing accelerated MRI techniques often struggle with two key challenges: (1) failure to effectively utilize inherent K-space prior information, leading to persistent aliasing artifacts from zero-filled inputs; and (2) contamination of target reconstruction quality by irrelevant information when employing multi-contrast fusion strategies. To overcome these challenges, we present MambaMDN, a dual-domain framework for multi-contrast MRI reconstruction. Our approach first employs fully-sampled reference K-space data to complete the undersampled target data, generating structurally aligned but modality-mixed inputs. Subsequently, we develop a Mamba-based modality disentanglement network to extract and remove reference-specific features from the mixed representation. Furthermore, we introduce an iterative refinement mechanism to progressively enhance reconstruction accuracy through repeated feature purification. Extensive experiments demonstrate that MambaMDN can significantly outperform existing multi-contrast reconstruction methods.

Problem

• Accelerates MRI scans by reconstructing undersampled data
• Reduces aliasing artifacts using K-space prior information
• Disentangles modality-specific features to improve reconstruction quality

Innovation

• Uses Mamba-based network for modality disentanglement
• Employs dual-domain framework with reference K-space completion
• Introduces iterative refinement for progressive feature purification