Denoising Simulated Low-Field MRI (70mT) using Denoising AutoEncoders (DAE) and Cycle-Consistent Generative Adversarial Network (Cycle-GAN)

In this work, a denoising Cycle-GAN is implemented to yield high-field, high resolution, high signal-to-noise ratio MRI images from simulated low-field, low resolution, low signal-to-noise MRI images. Resampling and additive Rician noise were used to simulate low-field MRI. Images were utilized to train a DAE and Cycle-GAN, with paired and unpaired cases, respectively. Both networks were evaluated using SSIM and PSNR image quality metrics. This work demonstrates the use of advanced machine learning to improve low-field MRI images that can outperform classical denoising autoencoders and does not require image pairs.