High-efficient Bloch simulation of magnetic resonance imaging sequences based on deep learning

Abstract

Objective: Bloch simulation constitutes an essential part of magnetic resonance imaging (MRI) development. However, even with the graphics processing unit (GPU) acceleration, the heavy computational load remains a major challenge, especially in large-scale, high-accuracy simulation scenarios. This work aims to develop a deep learning-based simulator to accelerate Bloch simulation. Approach: The simulator model, called Simu-Net, is based on an end-to-end convolutional neural network and is trained with synthetic data generated by traditional Bloch simulation. It uses dynamic convolution to fuse spatial and physical information with different dimensions and introduces position encoding templates to achieve position-specific labeling and overcome the receptive field limitation of the convolutional network. Main Results: Compared with mainstream GPU-based MRI simulation software, Simu-Net successfully accelerates simulations by hundreds of times in both traditional and advanced MRI pulse sequences. The accuracy and robustness of the proposed framework were verified qualitatively and quantitatively. Besides, the trained Simu-Net was applied to generate sufficient customized training samples for deep learning-based T2 mapping and comparable results to conventional methods were obtained in the human brain. Significance: As a proof-of-concept work, Simu-Net shows the potential to apply deep learning for rapidly approximating the forward physical process of MRI and may increase the efficiency of Bloch simulation for optimization of MRI pulse sequences and deep learning-based methods.

GPU-Based Parallel Processing Techniques for Enhanced Brain Magnetic Resonance Imaging Analysis: A Review of Recent Advances

“…According to this figure, the most used MRI sequence type is T1-weighted, and the sequence was used for different purposes such as segmentation [ 43 ] in 2023, image restoration in 2019 [ 52 ], reconstruction both in 2021 and 2022 [ 45 , 46 ], surface mapping in 2022 [ 44 ], and feature extraction in 2019 [ 54 ]. The second most used MRI sequence type is T2-weighted, and it was used for various tasks such as pulse sequence simulation in 2023 [ 42 ], reconstruction in 2021 [ 47 ], and simulation of a human head in 2021 [ 48 ]. The T1-contrast and the FLAIR sequences were used for the purpose of segmentation.…”

“…We basically examine key research publications about the integration of various high-performance computing technologies with medical image processing and analysis approaches in Table 2 . For instance, in the study by [ 42 ], the authors aimed to develop a deep learning-based simulation model called Simu-Net to speed up the process of Bloch simulation. In comparison with GPU-based MRI simulation software, the simulations in this study were successfully accelerated by Simu-Net.…”

FlexDTI: flexible diffusion gradient encoding scheme-based highly efficient diffusion tensor imaging using deep learning

“…Zhang et al (2021) utilized dynamic convolutional kernels to achieve the fusion of classification tasks and input information. Huang et al (2023) used dynamic convolution to fuse spatial and physical information with different dimensions and overcome the receptive field limitation of the convolutional network. In this work, we implement a dynamic-convolutionbased network to get the dynamical kernel parameters conditioned by the DW images and diffusion gradient direction information.…”

In vivo validation of Magnetic Resonance Spin Tomography in Time-domain

“…If, for example in a clinical setting, a fixed sequence (e.g. fixed flip angles at a fixed resolution) is used, a different network architecture may be utilized that sacrifices flexibility to gain additional runtime performance [80].…”