Abstract:PURPOSE:The common practice in acquiring the magnetic resonance (MR) images is to obtain two-dimensional (2D) slices at coarse locations while keeping the high in-plane resolution in order to ensure enough body coverage while shortening the MR scan time. The aim of this study is to propose a novel method to generate HR MR images from low-resolution MR images along the longitudinal direction. In order to address the difficulty of collecting paired low- and high-resolution MR images in clinical settings and to gain the advantage of parallel cycle consistent generative adversarial networks (CycleGANs) in synthesizing realistic medical images, we developed a parallel CycleGANs based method using a self-supervised strategy.METHODS AND MATERIALS:The proposed workflow consists of two parallely trained CycleGANs to independently predict the HR MR images in the two planes along the directions that are orthogonal to the longitudinal MR scan direction. Then, the final synthetic HR MR images are generated by fusing the two predicted images. MR images, including T1-weighted (T1), contrast enhanced T1-weighted (T1CE), T2-weighted (T2), and T2 Fluid Attenuated Inversion Recovery (FLAIR), of the multimodal brain tumor segmentation challenge 2020 (BraTS2020) dataset were processed to evaluate the proposed workflow along the cranial-caudal (CC), lateral, and anterior-posterior directions. Institutional collected MR images were also processed for evaluation of the proposed method. The performance of the proposed method was investigated via both qualitative and quantitative evaluations. Metrics of normalized mean absolute error (NMAE), peak signal-to-noise ratio (PSNR), edge keeping index (EKI), structural similarity index measurement (SSIM), information fidelity criterion (IFC), and visual information fidelity in pixel domain (VIFP) were calculated.RESULTS:It is shown that the proposed method can generate HR MR images visually indistinguishable from the ground truth in the investigations on the BraTS2020 dataset. In addition, the intensity profiles, difference images and SSIM maps can also confirm the feasibility of the proposed method for synthesizing HR MR images. Quantitative evaluations on the BraTS2020 dataset shows that the calculated metrics of synthetic HR MR images can all be enhanced for the T1, T1CE, T2, and FLAIR images. The enhancements in the numerical metrics over the low-resolution and bi-cubic interpolated MR images, as well as those genearted with a comparative deep learning method, are statistically significant. Qualitative evaluation of the synthetic HR MR images of the clinical collected dataset could also confirm the feasibility of the proposed method.CONCLUSIONS:The proposed method is feasible to synthesize HR MR images using self-supervised parallel CycleGANs, which can be expected to shorten MR acquisition time in clinical practices.

Generating Synthetic MR Spectroscopic Imaging Data with Generative Adversarial Networks to Train Machine Learning Models

Synthesized 7T MPRAGE From 3T MPRAGE Using Generative Adversarial Network and Validation in Clinical Brain Imaging: A Feasibility Study

Enhanced Magnetic Resonance Image Synthesis with Contrast-Aware Generative Adversarial Networks

7T MRI Synthesization from 3T Acquisitions

Metadata-Conditioned Generative Models to Synthesize Anatomically-Plausible 3D Brain MRIs

A Deep Learning Model for Multi-Domain MRI Synthesis Using Generative Adversarial Networks

Synthesizing High-Resolution Magnetic Resonance Imaging Using Parallel Cycle-Consistent Generative Adversarial Networks for Fast Magnetic Resonance Imaging.

Learning to Synthesize 7 T MRI from 3 T MRI with Few Data by Deformable Augmentation.

Generating Magnetic Resonance Spectroscopy Imaging Data of Brain Tumours from Linear, Non-linear and Deep Learning Models

Multi-sequence generative adversarial network: better generation for enhanced magnetic resonance imaging images

HybridGAN: Hybrid Generative Adversarial Networks for MR Image Synthesis

Atlas-Based Adaptive Hadamard-Encoded MR Spectroscopic Imaging at 3T

Synthetic Brain Images: Bridging the Gap in Brain Mapping With Generative Adversarial Model

Lesion Mask-based Simultaneous Synthesis of Anatomic and MolecularMR Images using a GAN

Multimodal MRI Synthesis Using Unified Generative Adversarial Networks.

MRI-only Based Synthetic CT Generation Using Dense Cycle Consistent Generative Adversarial Networks.

Generative Adversarial Training for MRA Image Synthesis Using Multi-Contrast MRI

Synthetic CT Generation from MRI Using 3D Improved Diffusion Model

Unpaired Multi-contrast MR Image Synthesis Using Generative Adversarial Networks.

Multimodal MR Image Synthesis Using Gradient Prior and Adversarial Learning