醫學影像的判讀既耗時又高度仰賴經驗，尤其是在核磁共振影像中產生的失真影像，而AI可自動執行偽影與運動失真的校正，從而縮短判讀流程並減少重複掃描。核磁共振掃描中約10–42% 的偽影源自受檢者顫抖、心跳或呼吸，其中約有20%需重做檢查，平均延長20–30分鐘並增加約15%成本，年均對每台儀器造成超過14萬美元的損失。而傳統校正倚賴繁瑣的人工作業，且需要動輒數十萬美元的硬體投入與冗長的後處理，難以兼顧影像品質與臨床效率。
本研究聚焦於核磁共振腦部成像因患者運動所致的偽影失真問題，採用公開T1腦部影像資料集，構建一套基於變分自編碼器的深度學習運動偽影校正模型，並整合了U型卷積神經網路的架構，為提升重建品質，引入了2.5維輸入策略，並設計雙權重線性插值的動態調整機制，以平衡重建與正則化損失項，並逐步增強變分自編碼器的KL項權重。
實驗結果表明，本模型在結構相似度指數、峰值訊噪比及歸一化均方根誤差等指標上，相較於運動偽影校正網路及深度密度先驗重建變分自編碼器均有顯著提升，能有效還原細節並抑制偽影，尤其在高度運動校正任務上效果更加優異，這證明了本研究模型的泛化能力與訓練穩定性。

Interpretation of medical images is both time consuming and highly dependent on expertise, especially for distorted MRI scans, while AI can automatically correct artifacts and motion distortions, shortening interpretation time and reducing repeat scans. In brain MRI, 10–42% of artifacts result from patient tremor, cardiac pulsation, or respiration, and about 20% of scans must be repeated, adding 20–30 minutes per exam and increasing costs by roughly 15%, at an annual loss exceeding $140,000 per scanner. Traditional correction depends on laborious manual adjustments and demands hundreds of thousands of dollars in hardware plus lengthy post-processing, making it difficult to balance image quality with clinical efficiency.
This study addresses motion-induced artifact distortion in brain MRI. Using a publicly available T1-weighted dataset, this work develops a deep learning-based motion-artifact correction model grounded in a Variational Autoencoder (VAE) and integrated with a U-shaped Convolutional Neural Network (U-Net). To enhance reconstruction quality, the study introduces a 2.5D multi-slice input strategy and designs a dynamic dual-weight linear-interpolation mechanism to balance reconstruction and regularization losses while progressively increasing the VAE's KL divergence weight. Experiments demonstrate that the proposed model significantly outperforms both the Motion Correction Network (MC-Net) and Deep Density Prior Reconstruction VAE (DDP Recon VAE) in Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), and Normalized Root Mean Square Error (NRMSE), effectively restoring fine details and suppressing artifacts, especially in high-motion correction tasks, thereby confirming its generalization capability and training stability.

[1] Andre, J. B., Bresnahan, B. W., Mossa-Basha, M., Hoff, M. N., Smith, C. P., Anzai, Y., & Cohen, W. A. （2015）. Toward quantifying the prevalence, severity, and cost associated with patient motion during clinical MR examinations. Journal of the American College of Radiology, 12（7）, 689-695
[2] Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4), 600-612.
[3] Hore, A., & Ziou, D. (2010, August). Image quality metrics: PSNR[3] vs. SSIM[2]. In 2010 20th international conference on pattern recognition (pp. 2366-2369). IEEE.
[4] Armstrong, J. S., & Collopy, F. (1992). Error measures for generalizing about forecasting methods: Empirical comparisons. International journal of forecasting, 8(1), 69-80.
[5] Legendre, A. M. (1806). Nouvelles méthodes pour la détermination des orbites des comètes: avec un supplément contenant divers perfectionnemens de ces méthodes et leur application aux deux comètes de 1805. Courcier.
[6] Laplace, P. S. (1774). Mémoire sur la probabilité de causes par les évenements. Mémoire de l'académie royale des sciences.
[7] Mathieu, M., Couprie, C., & LeCun, Y. (2015). Deep multi-scale video prediction beyond mean square error. arXiv preprint arXiv:1511.05440.
[8] Kullback, S., & Leibler, R. A. (1951). On information and sufficiency. The annals of mathematical statistics, 22(1), 79-86.
[9] Lauterbur, P. C. (1973). Image formation by induced local interactions: examples employing nuclear magnetic resonance. nature, 242(5394), 190-191.
[10] Akiba, T., Sano, S., Yanase, T., Ohta, T., & Koyama, M. (2019, July). Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining (pp. 2623-2631).
[11] Kingma, D. P., & Welling, M. （2013, December）. Auto-encoding variational bayes.
[12] Ronneberger, O., Fischer, P., & Brox, T. （2015）. U-Net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18 （pp. 234-241）. Springer international publishing.
[13] Setio, A. A. A., Ciompi, F., Litjens, G., Gerke, P., Jacobs, C., Van Riel, S. J., ... & Van Ginneken, B. （2016）. Pulmonary nodule detection in CT images: false positive reduction using multi-view convolutional networks. IEEE transactions on medical imaging, 35（5）, 1160-1169.
[14] Yang, J., Huang, X., He, Y., Xu, J., Yang, C., Xu, G., & Ni, B. （2021）. Reinventing 2d convolutions for 3d images. IEEE Journal of Biomedical and Health Informatics, 25（8）, 3009-3018.
[15] Tezcan, K. C., Baumgartner, C. F., Luechinger, R., Pruessmann, K. P., & Konukoglu, E. （2018）. MR image reconstruction using deep density priors. IEEE transactions on medical imaging, 38（7）, 1633-1642.
[16] Zhang, Y., Liu, M., Zhang, Z., & Dunson, D. （2024）. Motion-invariant variational autoencoding of brain structural connectomes. Imaging Neuroscience, 2, 1-27.
[17] Nárai, Á., Hermann, P., Auer, T., Kemenczky, P., Szalma, J., Homolya, I., ... & Vidnyánszky, Z. (2022). Movement-related artefacts (MR-ART) dataset of matched motion-corrupted and clean structural MRI brain scans. Scientific data, 9(1), 630.
[18] Loshchilov, I., & Hutter, F. (2017). Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101.
[19] LeCun, Y., Bottou, L., Orr, G. B., & Müller, K. R. (2002). Efficient backprop. In Neural networks: Tricks of the trade (pp. 9-50). Berlin, Heidelberg: Springer Berlin Heidelberg.
[20] Bowman, S. R., Vilnis, L., Vinyals, O., Dai, A. M., Jozefowicz, R., & Bengio, S. (2015). Generating sentences from a continuous space. arXiv preprint arXiv:1511.06349.