磁共振心臟影像能以多切面的方式重建3維空間的人體心臟影像，因此能提供足夠的資訊進行心臟功能的診斷。本研究以人工智慧的方式，訓練一個機器模型來完成心臟區塊分割的標籤繪製，同時使用包含時間序的四維磁共振心臟影像自動化計算心室射出率，提供有效率的心臟診斷方法。然而人工智慧有其極限，本研究提出以品質預測分數(QPS)，在資料預測期得以預知可能的模型失效。此外，本研究還探討了模型跨機構使用精準度不佳的問題，並提出改善方法。磁振造影成像因不同機構所使用的儀器或掃描方法不同，導致影像訊號分佈的差異，可能使得深度學習模型無法準確辨識分割心臟區塊。本研究提出數種資料增強法，來模擬因磁振物理形成之特有的影像偽影以及變異。實驗結果顯示出，資料增強法能提升預測的準確性，但依然無法達到移植前之精準表現。要改善這個問題，其中一個方法就是以使用機構之部份資料來進行模型微調。本研究提出以QPS來篩選微調用之訓練資料。實驗結果佐證，在標定移植機構資料之過程中，選擇較高之QPS進行模型微調，將得以用較少資料量來使得模型。

Magnetic resonance cardiac imaging can reconstruct human heart images in 3-dimensional space with multi-slices , thus providing sufficient information for the diagnosis of cardiac function. One of the cardiac quantitatively analyze: ejection fraction depends on the contours of the heart ventricle to calculate the parameters. However, manual delineation is time consuming and thus retards clinical diagnosis. Therefore, this study uses artificial intelligence to train the machine model which is used to predict the label of the heart segmentation, and the ejection fraction is automatically calculated by using a 4-dimensional cine magnetic resonance images containing time series to provide an efficient cardiac diagnostic method. This study also explores the relatively low accuracy when deploying models cross institutes and proposes improvement methods. For example, the magnetic resonance imaging would be different due to use the different MRI system or scanning methods by the different institutions, resulting in the difference image signal intensity distribution, which makes the model unable to accurately identify the heart and failed segmentation cardiac. So this study proposed an image processing and data augmentation method utilizing the unique proprieties of magnetic resonance imaging to improve prediction results of across cardiac image dataset of the accuracy. In addition, we propose to use the quality prediction score (QPS) to select datasets for fine-tuning the model. The results support that QPS provide an efficient way to include less datasets for the fine-tuning step during model deployment.

1. Schmidhuber, J., Deep learning in neural networks: An overview. Neural networks, 2015. 61: p. 85-117.
2. Russakovsky, O., et al., Imagenet large scale visual recognition challenge. International journal of computer vision, 2015. 115(3): p. 211-252.
3. Deng, J., et al. Imagenet: A large-scale hierarchical image database. in 2009 IEEE conference on computer vision and pattern recognition. 2009. Ieee.
4. Krizhevsky, A., I. Sutskever, and G.E. Hinton. Imagenet classification with deep convolutional neural networks. in Advances in neural information processing systems. 2012.
5. Shin, H.-C., et al., Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning. IEEE transactions on medical imaging, 2016. 35(5): p. 1285-1298.
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. Badrinarayanan, V., A. Kendall, and R. Cipolla, SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation. Ieee Transactions on Pattern Analysis and Machine Intelligence, 2017. 39(12): p. 2481-2495.
8. 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.
9. Yang, Y., et al., M-FCN: Effective fully convolutional network-based airplane detection framework. IEEE Geoscience and Remote Sensing Letters, 2017. 14(8): p. 1293-1297.
10. Bernard, O., et al., Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: Is the problem solved? IEEE transactions on medical imaging, 2018. 37(11): p. 2514-2525.
11. Ioffe, S. and C. Szegedy, Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167, 2015.
12. Glorot, X., A. Bordes, and Y. Bengio. Deep sparse rectifier neural networks. in Proceedings of the fourteenth international conference on artificial intelligence and statistics. 2011.
13. Scherer, D., A. Müller, and S. Behnke. Evaluation of pooling operations in convolutional architectures for object recognition. in International conference on artificial neural networks. 2010. Springer.
14. wikipedia.org. Bilinear Interpolation. 6 May 2019.
15. Nasr, G.E., E. Badr, and C. Joun. Cross entropy error function in neural networks: Forecasting gasoline demand. in FLAIRS conference. 2002.
16. Kingma, D.P. and J. Ba, Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
17. wikipedia.org. Softmax function. 26 June 2019.
18. Bai, W., et al., Automated cardiovascular magnetic resonance image analysis with fully convolutional networks. Journal of Cardiovascular Magnetic Resonance, 2018. 20(1): p. 65.