主動脈剝離佔所有急性主動脈症候群大約八至九成，其中典型主動脈剝離具有特徵如下:血管內膜剝離分隔出真腔與假腔。史丹佛分類將主動脈剝離分成甲、乙兩型，甲型定義為升主動脈有血管剝離，乙型為主動脈弓或降主動脈剝離但沒有影響升主動脈。主動脈剝離是可能致死的急症，若延遲診斷與治療，甲型在48小時內有超過五成的死亡率。因此，及早的偵測主動脈剝離並依據史丹佛分類給予後續治療，能提高病患存活率與降低致殘率。
因此，這篇研究的動機與目的為探討深度學習是否能利用電腦斷層血管影像有效偵測主動脈剝離，與進一步應用史丹佛分類法分類主動脈剝離種類。
在2015至2019年間，有130位病患接受電腦斷層主動脈血管攝影(57位甲型，43位乙型，和30位無剝離者)納入訓練與驗證集。二階層神經網路模型建置為第一階偵測主動脈剝離，第二階預測史丹佛分類法成甲、乙型，輸出類別機率值(介於0~1)。
使用五折交叉驗證法於訓練與測試集。在第一階段，平均骰子係數可達0.8766，切片基靈敏度、特異度、陽性預測值、陰性預測值、準確率分別為82.05% [95% CI: 74.32%, 89.79%], 99.07% [95% CI: 98.74%, 99.40%], 97.79% [95% CI: 96.96%, 98.61%], 91.61% [95% CI: 88.13%, 95.09%]，和 93.30% [95% CI: 90.55%, 96.06%]。在第二階段，切片基靈敏度、特異度、陽性預測值、陰性預測值、準確率分別為95.79% [95% CI: 93.07%, 98.52%], 97.29% [95% CI: 96.38%, 98.20%], 91.71% [95% CI: 89.26%, 94.15%], 98.67% [95% CI: 97.77%, 99.57%], and 97.09% [95% CI: 95.94%, 98.24%]。
這項研究的結果顯示，二階層神經網路模型可以準確偵測是否存在主動脈剝離與預測史丹佛分類，推測具有潛在用途於急性情況，幫助早期診斷典型主動脈剝離。

Classic aortic dissection presents with an intimal flap dividing the aorta into the false and true lumen, accounting for approximately 80% to 90% of acute aortic syndrome, is the most common manifestation. The Stanford classification is adopted worldwide for aortic dissection. Stanford type A is diagnosed if the ascending thoracic aorta is involved. Stanford type B is diagnosed when the ascending thoracic aorta is spared. The mortality is high to type A, and it owes over 50% mortality rate within 48 hours. Therefore the early diagnosis is important to avoid mortality and morbidity.
Deep-learning AD detection is well studied previously, but Stanford classification essential to guide management by a neural network is not evaluated. This study aimed to evaluate the feasibility of automatic Stanford classification of classic dissected aorta (AD) using a two-hierarchical convolutional neural network.
Between 2015 and 2019, 130 arterial phase series (57 type A, 43 type B and 30 negative cases) in computed tomography angiography (CTA) of the aorta were collected as the training and validation dataset. A 2-step hierarchical model was built, including the first step detecting AD and the second step predicting the probability (0-1) of Stanford types.
A 5-fold cross-validation was applied for training and validation. In the first step, the average Dice score for the attention U-net achieved 0.8766. The slice-based sensitivity, specificity, PPV, and NPV was 82.05% [95% CI: 74.32%, 89.79%], 99.07% [95% CI: 98.74%, 99.40%], 97.79% [95% CI: 96.96%, 98.61%], and 91.61% [95% CI: 88.13%, 95.09%], respectively. The slice-based accuracy was 93.30% [95% CI: 90.55%, 96.06%]. In the second step, the sliced-based sensitivity, specificity, PPV, NPV, and accuracy was 95.79% [95% CI: 93.07%, 98.52%], 97.29% [95% CI: 96.38%, 98.20%], 91.71% [95% CI: 89.26%, 94.15%], 98.67% [95% CI: 97.77%, 99.57%], and 97.09% [95% CI: 95.94%, 98.24%], respectively.
This study showed that this 2-step hierarchical model could detect and predict Stanford classification with high sensitivity, specificity and PPV, suggesting the potential application in an emergent setting to assist early diagnosis.

REFERENCES
[1] Baliga R. R. et al., "The Role of Imaging in Aortic Dissection and Related Syndromes," vol. 7, no. 4, pp. 406-424, 2014.
[2] Yu H.-Y., Chen Y.-S., Huang S.-C., Wang S.-S., and Lin F.-Y., "Late outcome of patients with aortic dissection: study of a national database," vol. 25, no. 5, pp. 683-690, 2004.
[3] Meszaros I. et al., "Epidemiology and clinicopathology of aortic dissection," vol. 117, no. 5, pp. 1271-1278, 2000.
[4] Clouse W. D. et al., "Acute aortic dissection: population-based incidence compared with degenerative aortic aneurysm rupture," in Mayo Clinic Proceedings, 2004, vol. 79, no. 2: Elsevier, pp. 176-180.
[5] Bossone E., Labounty T. M., and Eagle K. A., "Acute aortic syndromes: diagnosis and management, an update," European Heart Journal, vol. 39, no. 9, pp. 739-749d, 2018.
[6] McMahon M. A. and Squirrell C. A., "Multidetector CT of Aortic Dissection: A Pictorial Review," (in eng), Radiographics : a review publication of the Radiological Society of North America, Inc, vol. 30, no. 2, pp. 445-460, Mar 2010.
[7] Dohnert W., "Cardiovascular disorders: aortic dissection," Radiology review manual. 5th ed. Philadelphia, Pa: Lippincott Williams & Wilkins, pp. 607-609, 2005.
[8] Karmy-Jones R., Aldea G., and Boyle Jr E. M. J. C., "The Continuing Evolution in the Management," vol. 117, p. 1221, 2000.
[9] Yasaka K. and Abe O. J. P. m., "Deep learning and artificial intelligence in radiology: Current applications and future directions," vol. 15, no. 11, p. e1002707, 2018.
[10] Esteva A. et al., "Dermatologist-level classification of skin cancer with deep neural networks," vol. 542, no. 7639, pp. 115-118, 2017.
[11] Gulshan V. et al., "Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs," vol. 316, no. 22, pp. 2402-2410, 2016.
[12] Yasaka K., Akai H., Abe O., and Kiryu S. J. R., "Deep learning with convolutional neural network for differentiation of liver masses at dynamic contrast-enhanced CT: a preliminary study," vol. 286, no. 3, pp. 887-896, 2018.
[13] Dehghan E., Wang H., and Syeda-Mahmood T., "Automatic detection of aortic dissection in contrast-enhanced CT," in 2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017), 18-21 April 2017 2017, pp. 557-560, doi: 10.1109/ISBI.2017.7950582.
[14] Harris R. J. et al., "Classification of Aortic Dissection and Rupture on Post-contrast CT Images Using a Convolutional Neural Network," (in eng), Journal of digital imaging, Sep 12 2019.
[15] Krissian K., Carreira J. M., Esclarin J., and Maynar M., "Semi-automatic segmentation and detection of aorta dissection wall in MDCT angiography," (in eng), Medical image analysis, vol. 18, no. 1, pp. 83-102, Jan 2014.
[16] Xiaojie D., Meichen S., Jianming W., He Z., and Dandan C., "Segmentation of the Aortic Dissection from CT Images Based on Spatial Continuity Prior Model," in 2016 8th International Conference on Information Technology in Medicine and Education (ITME), 23-25 Dec. 2016 2016, pp. 275-280, doi: 10.1109/ITME.2016.0069.
[17] Yellapragada M. S., Xie Y., Graf B., Richmond D., Krishnan A., and Sitek A., "Deep learning based detection of acute aortic syndrome in contrast CT images," in 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI), 2020: IEEE, pp. 1474-1477.
[18] Li J., Cao L., Ge Y., Wang C., Meng B., and Guo W., "Multi-Task Deep Convolutional Neural Network for the Segmentation of Type B Aortic Dissection," 2018.
[19] Oktay O. et al., "Attention u-net: Learning where to look for the pancreas," 2018.
[20] Panchal A., Trivedi H., Rajput M., and Trivedi D., "Activity Recognition Using Temporal Features and Deep Bottleneck 3D-ResNeXt," Singapore, 2020: Springer Singapore, in Proceedings of First International Conference on Computing, Communications, and Cyber-Security (IC4S 2019), pp. 889-904.
[21] He K., Zhang X., Ren S., and Sun J., "Deep Residual Learning for Image Recognition," arXiv pre-print server, 2015-12-10 2015.
[22] Zou K. H. et al., "Statistical validation of image segmentation quality based on a spatial overlap index," (in eng), Acad Radiol, vol. 11, no. 2, pp. 178-189, 2004.
[23] Cheng J., Tian S., Yu L., Ma X., and Xing Y., "A deep learning algorithm using contrast-enhanced computed tomography (CT) images for segmentation and rapid automatic detection of aortic dissection," Biomedical Signal Processing and Control, vol. 62, p. 102145, 2020/09/01/ 2020.
[24] Tan Y., Tan L., Xiang X., Tang H., Qin J., and Pan W., "Automatic Detection of Aortic Dissection Based on Morphology and Deep Learning," Computers, Materials & Continua, vol. 62, no. 3, pp. 1201-1215, 2020.
[25] Singh S. P., Wang L., Gupta S., Goli H., Padmanabhan P., and Gulyás B. J. S., "3D deep learning on medical images: a review," vol. 20, no. 18, p. 5097, 2020.
[26] Köpüklü O., Gunduz A., Kose N., and Rigoll G., "Real-time Hand Gesture Detection and Classification Using Convolutional Neural Networks," in 2019 14th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2019), 14-18 May 2019 2019, pp. 1-8, doi: 10.1109/FG.2019.8756576.
[27] Hiratzka L. F. et al., "2010 ACCF/AHA/AATS/ACR/ASA/SCA/SCAI/SIR/STS/SVM Guidelines for the Diagnosis and Management of Patients With Thoracic Aortic Disease," vol. 121, no. 13, pp. e266-e369, 2010.
[28] Vardhanabhuti V. et al., "Recommendations for accurate CT diagnosis of suspected acute aortic syndrome (AAS)--on behalf of the British Society of Cardiovascular Imaging (BSCI)/British Society of Cardiovascular CT (BSCCT)," (in eng), Br J Radiol, vol. 89, no. 1061, pp. 20150705-20150705, 2016.
[29] Hata A. et al., "Deep learning algorithm for detection of aortic dissection on non-contrast-enhanced CT," European Radiology, vol. 31, no. 2, pp. 1151-1159, 2021/02/01 2021.