SRM-Net: Joint Sampling and Reconstruction and Mapping Network for Accelerated 3T Brain Multi-Parametric MR Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Multi-parametric magnetic resonance imaging (MRI) offers quantitative insights for medical diagnosis but suffers from long acquisition times.
• Existing accelerated MRI reconstruction methods show suboptimal performance due to sequential optimization and poor information utilization.

Purpose of the Study:

• To propose an integrated deep learning framework, SRM-Net, for accelerated multi-parametric MRI reconstruction.
• To enhance the efficiency and performance of generating multi-parametric maps from accelerated MRI data.

Main Methods:

• Developed SRM-Net, an end-to-end network with joint sampling, reconstruction, and mapping modules.
• Implemented an optimized sampling scheme and a spatio-temporal attention mechanism within a dual-domain unrolling framework.
• Utilized multi-layer perceptron for complex nonlinear mapping in the final module.

Main Results:

• SRM-Net achieved superior reconstruction of multi-parametric maps (T1, T2*, PD) for brain imaging on a 3T scanner.
• The method demonstrated improved utilization of intra- and inter-contrast information compared to state-of-the-art techniques.
• Promising intermediate weighted MR images were generated alongside the quantitative maps.

Conclusions:

• SRM-Net offers an effective all-in-one solution for accelerated multi-parametric MRI.
• The proposed framework significantly improves the quality of quantitative MRI maps and weighted images.
• SRM-Net holds potential for advancing medical diagnosis through faster and more accurate MRI.