Evaluation of motion artefact reduction depending on the artefacts' directions in head MRI using conditional generative adversarial networks

Abstract

Motion artefacts caused by the patient's body movements affect magnetic resonance imaging (MRI) accuracy. This study aimed to compare and evaluate the accuracy of motion artefacts correction using a conditional generative adversarial network (CGAN) with an autoencoder and U-net models. The training dataset consisted of motion artefacts generated through simulations. Motion artefacts occur in the phase encoding direction, which is set to either the horizontal or vertical direction of the image. To create T2-weighted axial images with simulated motion artefacts, 5500 head images were used in each direction. Of these data, 90% were used for training, while the remainder were used for the evaluation of image quality. Moreover, the validation data used in the model training consisted of 10% of the training dataset. The training data were divided into horizontal and vertical directions of motion artefact appearance, and the effect of combining this data with the training dataset was verified. The resulting corrected images were evaluated using structural image similarity (SSIM) and peak signal-to-noise ratio (PSNR), and the metrics were compared with the images without motion artefacts. The best improvements in the SSIM and PSNR were observed in the consistent condition in the direction of the occurrence of motion artefacts in the training and evaluation datasets. However, SSIM > 0.9 and PSNR > 29 dB were accomplished for the learning model with both image directions. The latter model exhibited the highest robustness for actual patient motion in head MRI images. Moreover, the image quality of the corrected image with the CGAN was the closest to that of the original image, while the improvement rates for SSIM and PSNR were approximately 26% and 7.7%, respectively. The CGAN model demonstrated a high image reproducibility, and the most significant model was the consistent condition of the learning model and the direction of the appearance of motion artefacts.

Figure 1

Overview of creating the simulated motion artefacts image in the magnetic resonance images.

Figure 2

(a) Conditional generative adversarial network (CGAN) framework, (b) network structure of the generator, and c discriminator. The training model consists of one generator and one discriminator with a conditional argument. To train the CGAN, the overall network’s performance is enhanced through networks acting bidirectionally with each other. A motion artefact correction image is generated by a network that maps images from a source domain (with motion artefact) to the target domain (artefacts correction image) based on the conditional ideal image.

Figure 3

Results of motion-corrected images by the autoencoder (AE), U-net, and conditional generative adversarial network (CGAN) models. (a) Simulated motion artefacts images and original magnetic resonance images, (b) Motion artefacts occurred in the horizontal direction, and (c) in the vertical direction. These models were trained using the image data; consistent with the direction of the motion, inconsistent with the direction of the motion, and with motion artefacts in the horizontal and vertical directions. Enlarged images of the region of interest, indicated in red a, are shown at the center of each image. For example, slices of the parietal image were shown on the right side of each image.

Figure 4

SSIM under training conditions consistent with the direction of the motion artefacts. * indicates p < 0.005, showing a significant difference to the result of CGAN, ** indicates p < 0.005, showing a significant difference to the results with corresponding motion artefacts image.

Figure 5

PSNR under training conditions consistent with the direction of the motion artefacts. * indicates p < 0.005, showing a significant difference to the result of CGAN, ** indicates p < 0.005, showing a significant difference to the results with corresponding motion artefacts image.