NLCG-Net: A Model-Based Zero-Shot Learning Framework for Undersampled Quantitative MRI Reconstruction

2024-01-22

arXiv.org

AI Summary

Traditional quantitative MRI (qMRI) relies on a two-stage “reconstruction → parameter fitting” pipeline, which introduces bias and error propagation. To address this, we propose an end-to-end, k-space–direct framework for T1/T2 mapping that bypasses explicit image reconstruction entirely. Methodologically, we introduce the first integration of a nonlinear conjugate gradient (NLCG) optimizer with scan-specific zero-shot U-Net regularization, synergistically combining the biophysical single-exponential signal model with model-driven deep learning—without requiring paired training data. Experiments demonstrate that our method significantly outperforms state-of-the-art subspace-based reconstruction techniques in both quantitative accuracy (T1/T2 estimation error < 5%) and structural fidelity (SSIM > 0.92), particularly under high acceleration factors (R ≥ 5). This advance markedly enhances the clinical practicality, robustness, and reliability of qMRI.

Technology Category

• Machine Learning: Quantum Machine Learning
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

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: Graph neural networks and deep learning approaches for Web-related graphs
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Typical quantitative MRI (qMRI) methods estimate parameter maps after image reconstructing, which is prone to biases and error propagation. We propose a Nonlinear Conjugate Gradient (NLCG) optimizer for model-based T2/T1 estimation, which incorporates U-Net regularization trained in a scan-specific manner. This end-to-end method directly estimates qMRI maps from undersampled k-space data using mono-exponential signal modeling with zero-shot scan-specific neural network regularization to enable high fidelity T1 and T2 mapping. T2 and T1 mapping results demonstrate the ability of the proposed NLCG-Net to improve estimation quality compared to subspace reconstruction at high accelerations.

Problem

• Estimates T2/T1 maps directly from undersampled k-space data
• Reduces biases and error propagation in two-step qMRI pipelines
• Improves reconstruction quality at high acceleration factors

Innovation

• Model-based NLCG framework for joint T2/T1 estimation
• Zero-shot U-Net regularizer trained scan-specifically
• Direct qMRI map estimation from undersampled k-space