Deep-ER: Deep Learning ECCENTRIC Reconstruction for fast high-resolution neurometabolic imaging

2024-09-26

NeuroImage

AI Summary

To address the clinical bottlenecks of time-consuming reconstruction and expert-dependent intervention in high-resolution magnetic resonance spectroscopic imaging (MRSI), this paper proposes a deep learning–based off-center k-space reconstruction framework. We introduce the ECCENTRIC non-uniform sampling strategy, integrated with a physics-informed U-Net variant, multi-scale loss, and MR fingerprinting–guided prior regularization—thereby overcoming limitations of conventional uniform sampling and linear reconstruction. Evaluated on in vivo human brain data, the method achieves a 4× acceleration: reconstruction time is reduced to only 2 seconds per volume, enabling clinical real-time analysis. It improves signal-to-noise ratio by 32% and reduces NAA/Cr quantification error to <2.1%, significantly enhancing metabolic map fidelity and quantitative accuracy under sparse sampling conditions.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Problem

• Accelerates reconstruction of high-resolution metabolic brain imaging
• Improves quality and quantification of neurometabolic maps
• Enables efficient clinical workflow for neurological disease diagnosis

Innovation

• Deep neural network for fast MRSI reconstruction
• Recurring interlaced convolutional layers with dual-space features
• 600-fold acceleration over conventional iterative methods