從影像中分割心臟的不同區域是確立心臟磁共振診斷的一項常見臨床任務，深度學習在此領域已展現出顯著的性能，如自動心臟診斷挑戰賽等。本論文通過數據增強技術訓練網絡，並證明該模型有助於更有效地學習心臟影像分割。我們進一步比較了2D U-net和3D U-net模型的分割性能，實驗結果表明2D U-net模型的分割性能優於3D U-net模型。除此之外，呼吸運動是心臟磁共振造影的挑戰之一，為了降低影像中的假影，我們使用影像處理方式生成的運動假影影像訓練模型，並在真實假影影像上測試，實驗結果證明能有效減少自由呼吸影像中的假影，且在分割任務上表現出更高的穩定性。總結來說，本研究探討模型在跨院資料上的分割及去噪的可靠性，實驗結果說明我們的模型在臨床實踐中具有可行性。

Segmenting heart subregions from cardiac magnetic resonance (CMR) images using deep learning is a clinical task for establishing diagnosis from CMR. Significant performance has also been demonstrated in previous competitions such as Automatic Cardiac Diagnosis Challenge. In this thesis, we trained the deep-learning network using data augmentation techniques and demonstrated that this model facilitates more effective learning of CMR image segmentation. We further compared the segmentation performance of the 2D U-net and 3D U-net models. The experimental results revealed that the segmentation performance of the 2D U-net model was superior. In addition, respiratory motion has been one of the challenges in acquiring CMR images. To address the presence of motion artifacts, we trained the model using artificially generated motion artifact images and evaluated it on real motion artifact images. The experimental results demonstrated its effectiveness in reducing the presence of artifacts in the free-breathing images, and the model exhibited increased stability in the segmentation task. In summary, this study investigated the reliability of the model for segmentation and denoising on cross-hospital data. The experimental results demonstrated the feasibility of our model in clinical practice.

[1] O. Huttin, M.-A. Petit, E. Bozec, R. Eschalier, Y. Juillière, F. Moulin, S. Lemoine, C. Selton-Suty, N. Sadoul, and D. Mandry, “Assessment of left ventricular ejection fraction calculation on long-axis views from cardiac magnetic resonance imaging in patients with acute myocardial infarction,” Medicine, vol. 94, no. 43, 2015.
[2] M. H. Nguyen, E. Abdelmaguid, J. Huang, S. Kenchareddy, D. Singla, L. Wilke, M. Bobar, E. D. Carruth, D. Uys, and I. Altintas, "Analytics pipeline for left ventricle segmentation and volume estimation on cardiac MRI using deep learning." pp. 305-306.
[3] C. Chen, C. Qin, H. Qiu, G. Tarroni, J. Duan, W. Bai, and D. Rueckert, “Deep learning for cardiac image segmentation: a review,” Frontiers in Cardiovascular Medicine, vol. 7, pp. 25, 2020.
[4] O. Bernard, A. Lalande, C. Zotti, F. Cervenansky, X. Yang, P.-A. Heng, I. Cetin, K. Lekadir, O. Camara, and M. A. G. Ballester, “Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: is the problem solved?,” IEEE transactions on medical imaging, vol. 37, no. 11, pp. 2514-2525, 2018.
[5] P. Peng, K. Lekadir, A. Gooya, L. Shao, S. E. Petersen, and A. F. Frangi, “A review of heart chamber segmentation for structural and functional analysis using cardiac magnetic resonance imaging,” Magnetic Resonance Materials in Physics, Biology and Medicine, vol. 29, pp. 155-195, 2016.
[6] S. Wang, C. Qin, N. Savioli, C. Chen, D. P. O’Regan, S. Cook, Y. Guo, D. Rueckert, and W. Bai, "Joint motion correction and super resolution for cardiac segmentation via latent optimisation." pp. 14-24.
[7] P. F. Ferreira, P. D. Gatehouse, R. H. Mohiaddin, and D. N. Firmin, “Cardiovascular magnetic resonance artefacts,” Journal of Cardiovascular Magnetic Resonance, vol. 15, no. 1, pp. 1-39, 2013.
[8] V. M. Campello, P. Gkontra, C. Izquierdo, C. Martin-Isla, A. Sojoudi, P. M. Full, K. Maier-Hein, Y. Zhang, Z. He, and J. Ma, “Multi-centre, multi-vendor and multi-disease cardiac segmentation: the M&Ms challenge,” IEEE Transactions on Medical Imaging, vol. 40, no. 12, pp. 3543-3554, 2021.
[9] C. Martín-Isla, V. M. Campello, C. Izquierdo, K. Kushibar, C. Sendra-Balcells, P. Gkontra, A. Sojoudi, M. J. Fulton, T. W. Arega, and K. Punithakumar, “Deep Learning Segmentation of the Right Ventricle in Cardiac MRI: The M&ms Challenge,” IEEE Journal of Biomedical and Health Informatics, 2023.
[10] O. Ronneberger, P. Fischer, and T. Brox, "U-net: Convolutional networks for biomedical image segmentation." pp. 234-241.
[11] C. Ren, N. Wang, and Y. Zhang, "Human Brain Hippocampus Segmentation Based on Improved U-net Model." pp. 7-11.
[12] D. P. Kingma, and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014.
[13] C. Shorten, and T. M. Khoshgoftaar, “A survey on image data augmentation for deep learning,” Journal of big data, vol. 6, no. 1, pp. 1-48, 2019.
[14] S. Yang, W. Xiao, M. Zhang, S. Guo, J. Zhao, and F. Shen, “Image data augmentation for deep learning: A survey,” arXiv preprint arXiv:2204.08610, 2022.
[15] B. Lorch, G. Vaillant, C. Baumgartner, W. Bai, D. Rueckert, and A. Maier, “Automated detection of motion artefacts in MR imaging using decision forests,” Journal of medical engineering, vol. 2017, 2017.
[16] I. Oksuz, “Brain MRI artefact detection and correction using convolutional neural networks,” Computer Methods and Programs in Biomedicine, vol. 199, pp. 105909, 2021.
[17] P. Y. Simard, D. Steinkraus, and J. C. Platt, "Best practices for convolutional neural networks applied to visual document analysis."