Autoencoder-Inspired Convolutional Network-Based Super-Resolution Method in MRI.

Seonyeong Park,H Michael Gach,Siyong Kim,Yuichi Motai,Suk Jin Lee

doi:10.1109/jtehm.2021.3076152

Abstract

Objective: To introduce an MRI in-plane resolution enhancement method that estimates High-Resolution (HR) MRIs from Low-Resolution (LR) MRIs. Method & Materials: Previous CNN-based MRI super-resolution methods cause loss of input image information due to the pooling layer. An Autoencoder-inspired Convolutional Network-based Super-resolution (ACNS) method was developed with the deconvolution layer that extrapolates the missing spatial information by the convolutional neural network-based nonlinear mapping between LR and HR features of MRI. Simulation experiments were conducted with virtual phantom images and thoracic MRIs from four volunteers. The Peak Signal-to-Noise Ratio (PSNR), Structure SIMilarity index (SSIM), Information Fidelity Criterion (IFC), and computational time were compared among: ACNS; Super-Resolution Convolutional Neural Network (SRCNN); Fast Super-Resolution Convolutional Neural Network (FSRCNN); Deeply-Recursive Convolutional Network (DRCN). Results: ACNS achieved comparable PSNR, SSIM, and IFC results to SRCNN, FSRCNN, and DRCN. However, the average computation speed of ACNS was 6, 4, and 35 times faster than SRCNN, FSRCNN, and DRCN, respectively under the computer setup used with the actual average computation time of 0.15 s per n}{}100times100n

Highlights

Magnetic Resonance Imaging (MRI) has superior soft tissue contrast versus x-ray-based imaging techniques such as Computed Tomography (CT) and cone beam CT [1]
Real-time four-dimensional (4D) MRI is being developed for MRI-guided Radiation Therapy (MRIgRT). 4D MRI typically suffers from low spatial resolution (e.g., ≥ 3.5 mm in-plane resolution) due to the constraints of k-space acquisition, temporal resolution, and system latency [2]
In the experiments, Autoencoder-inspired Convolutional Network-based Super-resolution (ACNS) achieved comparable Peak Signal-to-Noise Ratio (PSNR), Structure SIMilarity index (SSIM), and Information Fidelity Criterion (IFC) results to other Convolutional Neural Network (CNN)-based SR methods, and substantially reduced the computational time compared to the methods currently considered state-of-theart

Summary

Introduction

Magnetic Resonance Imaging (MRI) has superior soft tissue contrast versus x-ray-based imaging techniques such as Computed Tomography (CT) and cone beam CT [1]. MRIs can be acquired continuously without the risks of ionizing radiation. MRI-guided Radiation Therapy (MRIgRT) is preferred for treating tumors that are affected by motion, including lung tumors located near critical or radiosensitive organs i.e. organs at risk (OARs) such as the esophagus, heart, or major vessels [1]. A cine of a single image plane containing the tumor is acquired to gate radiation dose delivery in MRIgRT. Treatment gating is desired using the entire tumor volume and neighboring OARs to minimize errors associated with through-plane tumor motion. 4D MRI typically suffers from low spatial resolution (e.g., ≥ 3.5 mm in-plane resolution) due to the constraints of k-space acquisition, temporal resolution, and system latency [2]. New techniques are required to optimize the spatial and temporal resolution of 4D MRI in MRIgRT

Methods

Results

Discussion

Conclusion