Simultaneous reduction of noise and motion artifacts in brain MRI using deep learning.

Motivation: The aim is to contribute to diagnosis by simultaneously reducing motion artifacts and noise in head MRI images using deep learning. Goal(s): The goal is to achieve high motion artifact and noise reduction in T1W, T2W, and FLAIR images, independent of artifact and noise levels. Approach: Simulation was used to create an image containing motion artifacts and noise, and deep learning was used to evaluate the removal effect. Results: The average SSIM between the ground troth and the input image was 0.72, and the SSIM between the ground troth and the output image using this method was 0.95, showing a high improvement effect. Impact: By 36,000 pairs of training data, we were able to increase the accuracy of the learning process. The advantage of this method is that it is post-processing and can be used regardless of the equipment or imaging method.