JotlasNet: Joint tensor low-rank and attention-based sparse unrolling network for accelerating dynamic MRI.

2025-01-01

Magnetic Resonance Imaging

AI Summary

Existing dynamic MRI reconstruction methods predominantly rely on matrix-based low-rank priors, neglecting intrinsic tensor structures, and employ global thresholding for sparsity regularization—limiting adaptability and modeling capacity. To address these limitations, we propose JotlasNet, a novel deep unrolling network that, for the first time, jointly integrates tensor low-rank decomposition with channel-spatial dual-attention sparse regularization within a learnable physics-informed iterative reconstruction framework. This design enables synergistic data-driven and model-driven adaptive reconstruction. Quantitatively, JotlasNet achieves PSNR improvements of 2.3–3.1 dB over state-of-the-art methods across multiple dynamic MRI benchmarks, while maintaining sub-150 ms inference time per frame—meeting real-time clinical requirements.

Technology Category

• Machine Learning: Matrix & Tensor Methods
• Computer Vision: Low Level & Physics-based Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Problem

• Dynamic MRI reconstruction enhancement
• Tensor low-rank prior utilization
• Attention-based sparse thresholding optimization

Innovation

• Utilizes tensor low-rank priors
• Employs attention-based sparse thresholds
• Features simple parallel structure