主動脈剝離是致死率很高的一個疾病，因此立即的診斷、發現、與治療相當的重要。主動脈剝離的診斷主要是透過電腦斷層血管攝影(CT Angiograph, CTA)檢查，掃描流程共需二組(有/無施打顯影劑)影像，藉由兩者的差異，可確定有無剝離的病兆。然而由於電腦斷層血管攝影必須施打顯影劑並且二次CT掃描會有過多輻射劑量的風險。因此本研究希望結合深度學習技術直接透過未施打顯影劑的影像便能有效地預測有無主動脈剝離的徵兆。這項研究回溯式的收集南部某醫學中心，從2012年9月到2019年12月，接受主動脈CTA檢查的患者，共 1296人次。我們總共納入778位病人(包含有和無主動脈剝離的受檢者)，隨機分為訓練組、驗證組以及測試組三個資料集。並且建構了一個用於預測主動脈剝離的卷積神經網路。研究結果顯示在訓練組與驗證組中，卷積神經網路預測的準確率分別為0.977與0.925，損失率則分別為0.167與0.264。在測試組上，卷積神經網路預測的靈敏度為0.97，特異度0.90。並且在接收者操作特徵曲線分析中，曲線下面積可高達0.971。綜上所述，本研究的結果初步顯示應用卷積神經網路在無需施打顯影劑的CT影像上能夠準確地預測出主動脈剝離。未來我們將進一步將評估此項技術並且與臨床醫師的診斷進行更深入的比較。

Aortic dissection is a disease with a high fatality rate. Therefore, immediate diagnosis, detection, and treatment is crucial. Computer Tomography Angiograph (CTA) is the investigation of choice for aortic dissection. Entire scan processes of CAT require two sets of images (with/without contrast media). Based on the difference between the two images, one can determine whether there is a sign of dissection. However, CT angiography must be administered with contrast agent and there is a risk of excessive radiation dose due to the requirement of dual CT scans. Therefore, the study aims to effectively predict the aortic dissection merely from the non-contrast CT images via deep learning. This study retrospectively collected 1,296 patients who underwent aortic CTA examinations from a medical center in the south Taiwan from September 2012 to December 2019. We included a total of 778 patients (including subjects with and without aortic dissection diagnosed), which were randomly divided into three data sets: training group, validation group, and test group. A convolutional neural network (CNN) for predicting aortic dissection was then established. Results showed that the accuracy of CNN-prediction was 0.977 for the training group and 0.925 for the validation group, while the loss rate was 0.167 and 0.264, respectively. For the testing group, a sensitivity of 0.97 and a specificity of 0.90 can be achieved using CNN method. The area under the ROC curve be as high as 0.971. In summary, the preliminary results show the promise for predicting aortic dissection on non-contrast CT images using convolutional neural networks. We will further evaluate this technology and make an in-depth comparison with the diagnosis of clinicians in the future.

A. Hata, M. Yanagawa, K. Yamagata, Y. Suzuki, S. Kido, A. Kawata, et
al. 2021 Deep learning algorithm for detection of aortic
dissection on non-contrast-enhanced CT European Radiology Vol. 31
Issue 2 Pages 1151-1159
Bae K T 2010 Intravenous contrast medium administration and scan
timing at CT: considerations and approaches Radiology 256 32-61
Beckett K R, Moriarity A K and Langer J M 2015 Safe use of contrast
media: what the radiologist needs to know Radiographics 35 1738-50
Chartrand G, Cheng P M, Vorontsov E, Drozdzal M, Turcotte S, Pal C J,
Kadoury S and Tang A 2017 Deep learning: a primer for radiologists Radiographics 37 2113-31
Dehghan E, Wang H and Syeda-Mahmood T 2017 IEEE 14th
International Symposium on Biomedical Imaging (ISBI 2017),2017), vol. Series): IEEE) pp 557-60
Huo D, Kou B, Zhou Z and Lv M 2019 A machine learning model to
classify aortic dissection patients in the early diagnosis phase Scientific reports 9 1-8
Ketkar N and Santana E 2017 Deep learning with python vol 1: Springer)
pp 72
Krizhevsky A, Sutskever I and Hinton G E 2017 ImageNet classification
with deep convolutional neural networks Communications of the
ACM 60 84-90
Lecturio 2021 Computed Tomography(CT)|Concise Medical Knowledge
https://www.lecturio.com/concepts/computed-tomography-ct/
Liu L, Zhang C, Zhang G, Gao Y, Luo J, Zhang W, Li Y and Mu Y 2020
A study of aortic dissection screening method based on multiple
machine learning models Journal of thoracic disease 12 605
Lu L, Zheng Y, Carneiro G and Yang L 2017 Deep learning and
convolutional neural networks for medical image computing
Advances in Computer Vision and Pattern Recognition 10 978 -3
pp 48
Lu J-T, Brooks R, Hahn S, Chen J, Buch V, Kotecha G, Andriole K P,
Ghoshhajra B, Pinto J and Vozila P International Conference on
Medical Image Computing and Computer-Assisted
Intervention,2019), vol. Series): Springer) pp 723-31
Manaswi N K, Manaswi N K and John S 2018 Deep learning with
applications using python: Springer) pp 95
McMahon M A and Squirrell C A 2010 Multidetector CT of aortic
dissection: a pictorial review Radiographics 30 445-60
Norris D J 2017 Beginning artificial intelligence with the Raspberry Pi:
Springer) pp 42,213
Pattanayak S, Pattanayak and John S 2017 Pro deep learning with
tensorflow: Springer) pp 90,102,106.
Selvaraju R R, Cogswell M, Das A, Vedantam R, Parikh D and Batra D
Proceedings of the IEEE international conference on computer
vision,2017), vol. Series) pp 618-26
Swamynathan M 2017 Mastering machine learning with python in six
steps: A practical implementation guide to predictive data analytics
using python: Apress) pp 65
Tan Y 2020 Automatic Detection of Aortic Dissection Based on
Morphology and Deep Learning Computers, Materials & Continua
62 1201-15
Thomsen H S 2014 Management of Acute Adverse Reactions to
Contrast Media: Springer) pp 63-9
Trentin C, Faggiano E, Conti M and Auricchio F 2015 9th International
Symposium on Image and Signal Processing and Analysis (ISPA),2015), vol. Series): IEEE) pp 2,6,8.9