[2E1-GS-13] AI application: Medical application (2)

[2E1-GS-13-01] Classification of Functionally Significant Coronary Artery Stenosis using LSTM

〇Reika Kosuda¹, Yuki Onagi², Joji Ota³, Manami Takahashi⁴, Hiroyuki Takaoka⁴, Hajime Yokota⁵, Takuro Horikoshi⁶, Yasukuni Mori⁷, Hiroki Suyari⁷ (1. Chiba University Faculty of Engineering, 2. Graduate School of Chiba University, 3. Chiba University Hospital Department of Radiology, 4. Chiba University Hospital Department of Cardiovascular, 5. Chiba University Graduate School of Medicine, 6. Chiba University Hospital Department of Radiology , 7. Chiba University Graduate School of Engineering)

Keywords:LSTM, CNN, FFR, Automated diagnosis support system

Coronary artery stenosis is one of the main causes of heart disease, which is the second leading cause of death in Japan. CT is used for detection of coronary artery stenosis, and coronary angiography (CAG) is necessary when coronary artery stenosis is suspected. However, CAG is invasive and it is convenient to detect functionally significant coronary artery stenosis on CT, because unnecessary CAG can be avoided.Therefore, we propose a method to identify functionally significant coronary artery stenosis, using a deep learning model of CT images. Our model is based on convolutional neural networks and LSTM. Consecutive 150 short axial images of each coronary artery are used. We regard them as time series data and fractional flow reserve (FFR), which is used as an index of functionally significant stenosis, was evaluated.The results show that FFR can be estimated only from CT images although the recognition accuracy is not sufficient.