Retrospective motion correction for preclinical/clinical MRI based on conditional GAN with entropy loss.

Multi-shot scan MRI acquisition is inherently sensitive to motion, and motion artifact reduction is essential for improving the image quality in MRI. This work proposes and validates a new end-to-end motion correction method for the multi-shot sequence that incorporates the conditional Generative Adversarial Network with Minimum Entropy of MRI images (cGANME). The cGANME network contains an encoder-decoder generator to obtain the motion-corrected images and a PatchGAN discriminator to classify the image as either real (motion-free) or fake (motion-corrected). The entropy of the images is set as one loss item in the cGAN's loss as the entropy increases monotonically with the motion artifacts. The ablation experiment of the different weights of the entropy loss is performed to evaluate the function of entropy loss. The preclinical dataset is acquired with the FSE pulse sequence on a 7.0T scanner. After the simulation, we have 10080/2880/1440 slices for training, testing, and validating, respectively. The clinical data set is downloaded from the Human Connection Project website, 11300/3500/2000 slices are used for training, testing, and validating after simulation. Extensive experiments for different motion patterns, motion levels, and protocol parameters demonstrate that cGANME outperforms traditional and some state-of-art deep learning-based methods. In addition, we test cGANME on in vivo awake rats and mitigate the motion artifacts, indicating the model has a certain generalization ability.