肺部微灌流可以透過動態對比增強 MRI 來做定量評估分析，通常用手動圈
選肺部的 ROI，在必須排除靜脈組織和大血管的情況下，是這很耗時且非常依賴
操作員的技術。有研究使用自動切割的方式，通過定義相對血容量或相關係數與
動脈輸入函數的標準。然而，使用這樣的標準將不會選擇具有流動不足或改變的
曲線形狀的區域。深度學習是一種新技術，我們這次主要是運用 SegNet 神經網
路架構，其網路架構是由英國劍橋大學在 2016 年所提出，並且被證明對於各種
影像的切割和辨識是非常有用的。因此，我們將嘗試使用深度學習來實現肺部的
自動化切割。
本研究將藉由四種肺部的微灌流造影資料集，包括打藥前、最大值、平均值、
最大值對應的時間值等影像，使用 Dice coefficient 和計算肺部微灌流參數來評估
被切割後的位置，並且嘗試輸入不同的影像組合和改變網路架構的參數值進行訓
練來達到肺部切割的最佳化。在結論的部分，使用深度學習可以提供強大的自動
化肺部切割並且此模型可以應用在正常人和病患身上。

Pulmonary perfusion can be assessed quantitatively by dynamic contrast
enhanced MRI. ROIs of the lungs are usually manually selected to exclude static tissue
and large vessels. However, it is time-consuming and operator-dependent. There are
studies using automatic segmentation by defining criteria of relative blood volume or
correlation coefficient with arterial input function. However, regions with flow deficit
or changed curve shape will not be selected using such criteria. Deep learning is a novel
technique and proved to be very useful for segmentation for various applications.
Therefore, we try to use deep learning to achieve automatic lung segmentation.
This study uses four types of data from perfusion imaging, including baseline,
maximum, mean and time to peak. Dice coefficient and perfusion parameters were used
to evaluate the segmented ROIs. We have tried different combination of data types to
train deep learning process. We have also tried to change model’s parameters and
preprocessing of perfusion data to optimize the automatic lung segmentation. In
conclusion, deep learning can provide robust segmentation of the lung, and hope that
can be applied to normal people and patients.

1. 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.
2. Chen, L.-C., et al., Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs. IEEE transactions on pattern analysis and machine intelligence, 2018. 40(4): p. 834-848.
3. Cireşan, D., U. Meier, and J. Schmidhuber, Multi-column deep neural networks for image classification. arXiv preprint arXiv:1202.2745, 2012.
4. He, K., et al. Deep residual learning for image recognition. in Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
5. Hua, K.-L., et al., Computer-aided classification of lung nodules on computed tomography images via deep learning technique. OncoTargets and therapy, 2015. 8.
6. Kingma, D.P. and J. Ba, Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
7. Krizhevsky, A., I. Sutskever, and G.E. Hinton. Imagenet classification with deep convolutional neural networks. in Advances in neural information processing systems. 2012.
8. 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.
9. Simonyan, K. and A. Zisserman, Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014.
10. 楊采玲, 以深度學習法分析大腦磁振影像: 腦腫瘤之自動分區. 2018.
11. 歐育年, 基於深度學習演算法的全自動心臟影像多層分析系統. 2018.
12. 沈宗翰, 肺部微灌流磁振影像之全自動影像分析系統. 2007.
13. Detre, J.A., et al., Perfusion imaging. Magnetic resonance in medicine, 1992. 23(1): p. 37-45.
14. Fink, C., et al. Quantitative analysis of pulmonary perfusion using time-resolved parallel 3D MRI-initial results. in RöFo-Fortschritte auf dem Gebiet der Röntgenstrahlen und der bildgebenden Verfahren. 2004. © Georg Thieme Verlag Stuttgart· New York.
15. Ohno, Y., et al., Primary pulmonary hypertension: 3D dynamic perfusion MRI for quantitative analysis of regional pulmonary perfusion. American Journal of Roentgenology, 2007. 188(1): p. 48-56.
16. Ng, C.S., et al., Reproducibility of perfusion parameters in dynamic contrast-enhanced MRI of lung and liver tumors: effect on estimates of patient sample size in clinical trials and on individual patient responses. American Journal of Roentgenology, 2010. 194(2): p. W134-W140.
17. Ingrisch, M., et al., Quantitative Pulmonary Perfusion Magnetic Resonance Imaging: Influence of Temporal Resolution and Signal-to-Noise Ratio. Investigative Radiology, 2010. 45(1): p. 7-14.