MAUDGAN: Motion Artifact Unsupervised Disentanglement Generative Adversarial Network of Multicenter MRI Data with Different Brain tumors

• DOI: 10.1101/2023.03.06.23285299
• Publication Date: 2023-03-08T17:55:17Z
• Authors: Mojtaba Safari; Ali Fatemi; Louis Archambault
• Keywords: Artifact (error); Similarity (geometry); Generative adversarial network

Abstract Purpose This study proposed a novel retrospective motion reduction method named artifact unsupervised disentanglement generative adversarial network (MAUDGAN) that reduces the artifacts from brain images with tumors and metastases. The MAUDGAN was trained using mutlimodal multicenter 3D T1-Gd T2-fluid attenuated inversion recovery MRI images. Approach different levels were simulated in k -space for consisted of two generators, discriminators feature extractor networks constructed residual blocks. generators map content space to vice-versa. On other hand, attempted discriminate codes learn motion-free motion-corrupted spaces. Results We compared CycleGAN Pix2pix-GAN. Qualitatively, could remove highest level soft-tissue contrasts without adding spatial frequency distortions. Quantitatively, we reported six metrics including normalized mean squared error (NMSE), structural similarity index (SSIM), multi-scale (MS-SSIM), peak signal-to-noise ratio (PSNR), visual information fidelity (VIF), gradient magnitude deviation (MS-GMSD). got lowest NMSE MS-GMSD. average, reconstructed SSIM, PSNR, VIF values comparable MS-SSIM values. Conclusions can disentangle dataset under multimodal framework. will improve automatic manual post-processing algorithms auto-segmentations, registrations, contouring guided therapies such as radiotherapy surgery.