本研究以人工智慧的方式，訓練用於完成心臟區塊分割的模型，尤其是針對心肌的詳細分割為目的，而在研究結果顯示，最佳的模型切割方式為將心臟不僅僅是切割出左心室與右心室的區塊，亦將心肌分割出心室中膈以及非心室中膈兩個區塊，使該模型不僅能夠提供足夠的資訊進行心臟功能的診斷（例如射血分率、心室容積等），亦能協助診斷罕見心臟疾病的特定症狀。然而人工智慧的模型會受到訓練資料的內容而有所限制，因此本研究亦探討模型在跨院資料上的精準度，並討論模型的預測結果以及預期的結果之間產生差異的原因。在實驗結果顯示，本研究所使用高同質性的資料集所訓練出來的模型，對於預測跨院資料上的預測能力較佳；反之，本研究所使用資料較為多樣且複雜的資料集，其必須經過重新資料分配，使其訓練資料的數量增加，再對其做訓練所訓練出來的模型，對於預測跨院資料的預測準確度能夠大幅的提升。

In this study, we used artificial intelligence to train a model to achieve the segmentation of the cardiac magnetic resonance images, especially for the detailed segmentation of the myocardium. We used three experiments to assess the effects of the cardiac labeling methods and the selection of training datasets. The experimental results showed that the best labeling method of the model was not only to segment the left ventricle and right ventricle from the cardiac MRI image but also to segment the myocardial segment into two blocks, septum, and non-septum. These results can be regarded as not only providing sufficient information for the diagnosis of cardiac function, such as ejection fraction or ventricular volume but also assisting in the diagnosis of symptoms of rare heart diseases. In addition, this study also discussed the accuracy of the model on the multi-site data and the potential causes of the discrepancy between the model results and the expected results.

1. Hammouda, K., et al., A new framework for performing cardiac Strain Analysis from cine MRi imaging in Mice. 2020. 10(1): p. 1-15.
2. Huang, H.-H., et al., Automatic regional analysis of myocardial native T1 values: left ventricle segmentation and AHA parcellations. 2018. 34(1): p. 131-140.
3. Cerqueira Manuel, D., et al., Standardized Myocardial Segmentation and Nomenclature for Tomographic Imaging of the Heart. Circulation, 2002. 105(4): p. 539-542.
4. Chen, C., et al., Deep learning for cardiac image segmentation: A review. 2020. 7: p. 25.
5. O'Shea, K. and R. Nash, An Introduction to Convolutional Neural Networks. ArXiv e-prints, 2015.
6. Long, J., E. Shelhamer, and T. Darrell. Fully convolutional networks for semantic segmentation. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2015.
7. Liu, X., et al., A Review of Deep-Learning-Based Medical Image Segmentation Methods. 2021. 13(3): p. 1224.
8. Noh, H., S. Hong, and B. Han. Learning deconvolution network for semantic segmentation. in Proceedings of the IEEE international conference on computer vision. 2015.
9. Goel, N., A. Yadav, and B.M. Singh. Medical image processing: a review. in 2016 Second International Innovative Applications of Computational Intelligence on Power, Energy and Controls with their Impact on Humanity (CIPECH). 2016. IEEE.
10. Taghanaki, S.A., et al., Deep semantic segmentation of natural and medical images: A review. 2021. 54(1): p. 137-178.
11. Ronneberger, O., P. Fischer, and T. Brox. U-net: Convolutional networks for biomedical image segmentation. in International Conference on Medical image computing and computer-assisted intervention. 2015. Springer.
12. Tao, Q., et al., Deep learning–based method for fully automatic quantification of left ventricle function from cine MR images: a multivendor, multicenter study. 2019. 290(1): p. 81-88.
13. Zhou, X.-Y., G.-Z.J.I.R. Yang, and A. Letters, Normalization in training U-Net for 2-D biomedical semantic segmentation. 2019. 4(2): p. 1792-1799.
14. Bernard, O., et al., Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: is the problem solved? 2018. 37(11): p. 2514-2525.