Multi-Resolution Parallel Magnetic Resonance Image Reconstruction in Mobile Computing-Based IoT.

Yan Chen,Qinglin Zhao,Xiping Hu,Bin Hu

DOI: https://doi.org/10.1109/ACCESS.2019.2894694

IF: 3.9

2019-01-01

IEEE Access

Abstract:In the mobile computing-based Internet of Things, the computational complexity of applications is constrained by the capacity of the user equipment. In order to reduce the computational complexity of compressed sensing (CS)-based magnetic resonance image (MRI) reconstruction algorithms, we propose a novel multi-resolution-based parallel MRI reconstruction framework in this paper. We break down CS-based MRI reconstruction problem into four independent low-resolution image reconstruction sub-problems. Compared with the original problem, each sub-problem has a lower computational complexity. Assigned to four cores of the central processing unit (CPU), the sub-problems are solved simultaneously, and therefore the MRI reconstruction is accelerated. The combination of reconstructed low-resolution images achieves high-resolution image reconstruction. The proposed framework is applicable to the state-of-the-art CS-based MRI reconstruction algorithms to compute low-resolution images and involves multi-resolution processing. Compared with conventional serial computing, the proposed MRI reconstruction framework speeds at least four times up. Therefore, the parallel computation framework is especially suitable for widely used mobile devices with lower computational capability than workstations. To validate and evaluate the proposed scheme, when selecting the outstanding MRI reconstructing algorithm fast dictionary learning method on classified patches for numerical simulation, peak-signal-to-noise-ratio values of parallel reconstruction results are at least 0.929 dB higher than that of serial computation reconstruction results calculated by classical MRI reconstruction algorithm.