Retrospective motion correction for preclinical/clinical magnetic resonance imaging based on a conditional generative adversarial network with entropy loss

Abstract

Multishot scan magnetic resonance imaging (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 multishot sequence that incorporates a conditional generative adversarial network with minimum entropy (cGANME) of MR images. The cGANME contains an encoder‐decoder generator to obtain 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. An ablation experiment of the different weights of entropy loss was performed to evaluate the function of entropy loss. The preclinical dataset was acquired with a fast spin echo pulse sequence on a 7.0‐T scanner. After the simulation, we had 10,080/2880/1440 slices for training, testing, and validating, respectively. The clinical dataset was downloaded from the Human Connection Project website, and 11,300/3500/2000 slices were used for training, testing, and validating after simulation, respectively. Extensive experiments for different motion patterns, motion levels, and protocol parameters demonstrate that cGANME outperforms traditional and some state‐of‐the‐art, deep learning‐based methods. In addition, we tested cGANME on in vivo awake rats and mitigated the motion artifacts, indicating that the model has some generalizability.

Correction of Arterial-Phase Motion Artifacts in Gadoxetic Acid-Enhanced Liver MRI Using an Innovative Unsupervised Network

“…Due to the challenges posed by the lack of ground truth data, supervised learning approaches are seldom practical for the motion correction of gadoxetic acid-enhanced MR images [16][17][18]29]. As a result, the current research focus has shifted towards unsupervised approaches.…”

“…Deep learning has emerged as a powerful tool in the field of MRI motion correction, offering a solution to address convergence issues often associated with retrospective techniques, as mentioned previously [16,18,28]. Early deep-learning models heavily relied on paired motion-free images for supervised learning, despite their proficiency in artifact correction [16][17][18]29]. Consequently, the feasibility of these supervised approaches diminishes due to the inherent challenge of acquiring paired motion-free images in clinics, especially in the context of enhanced MRI scans.…”

“…Afterwards, one notable breakthrough is the introduction of the Cycle-Consistent General Adversarial Network (CycleGAN) [20]. CycleGAN represents a significant advancement in the realm of motion artifact correction for liver MRI examinations [20,29]. A pivotal innovation in motion artifact correction using CycleGAN is the introduction of a new non-adversarial loss named cycle-consistency loss [20].…”

“…This loss function plays a crucial role in preserving vital image information and mitigating the risk of information loss during the image translation process. It ensures that the translation process from the motioncorrupted domain to the motion-free domain and back remains consistent, thus bolstering the correction of motion artifacts [20,29]. As a result, CycleGAN alleviates the need for paired motion-free and motion-corrupted images in clinical settings [20,29].…”

“…It ensures that the translation process from the motioncorrupted domain to the motion-free domain and back remains consistent, thus bolstering the correction of motion artifacts [20,29]. As a result, CycleGAN alleviates the need for paired motion-free and motion-corrupted images in clinical settings [20,29]. Furthermore, certain unsupervised methods (e.g., Cycle-MedGAN, etc.…”