Coronary artery disease, which is mostly caused by atherosclerotic narrowing of the coronary artery lumen, is a leading cause of death. Coronary angiography is the standard method to estimate the severity of coronary artery stenosis, but is frequently limited by intra- and inter-observer variations. We propose a deep-learning algorithm that automatically recognizes stenosis in coronary angiographic images. The proposed method consists of key frame detection, deep learning model training for classification of stenosis on each key frame, and visualization of the possible location of the stenosis. Firstly, we propose an algorithm that automatically extracts key frames essential for diagnosis from 452 right coronary artery angiography movie clips. Our deep learning model is then trained with image-level annotations to classify the areas narrowed by over 50 %. To make the model focus on the salient features, we apply a self-attention mechanism. The stenotic locations are visualized using the activated area of feature maps with gradient-weighted class activation mapping. The automatically detected key frame was very close to the manually selected key frame (average distance (1.70 ± 0.12) frame per clip). The model was trained with key frames on internal datasets, and validated with internal and external datasets. Our training method achieved high frame-wise area-under-the-curve of 0.971, frame-wise accuracy of 0.934, and clip-wise accuracy of 0.965 in the average values of cross-validation evaluations. The external validation results showed high performances with the mean frame-wise area-under-the-curve of (0.925 and 0.956) in the single and ensemble model, respectively. Heat map visualization shows the location for different types of stenosis in both internal and external data sets. With the self-attention mechanism, the stenosis could be precisely localized, which helps to accurately classify the stenosis by type. Our automated classification algorithm could recognize and localize coronary artery stenosis highly accurately. Our approach might provide the basis for a screening and assistant tool for the interpretation of coronary angiography.

Background—: We developed a model that predicts difficulty of percutaneous coronary intervention for coronary chronic total occlusion (CTO) using coronary computed tomographic angiography. Methods and Results—: A total of 684 CTO lesions with preprocedural computed tomographic angiography were enrolled from 4 centers. Data were randomly divided into derivation and validation datasets at 2:1 ratio. The end point was successful guidewire crossing ≤30 minutes, which was met in 50%. The KCCT (Korean Multicenter CTO CT Registry) score was developed based on independent predictors identified by multivariable analysis, which were proximal blunt entry, proximal side branch, bending, occlusion length ≥15 mm, severe calcification, whole luminal calcification, reattempt, and ≥12 months or unknown duration of occlusion. The KCCT score was compared with the other prediction scores, including angiography-based J-CTO, PROGRESS-CTO, CL-score, and CT-based CT-RECTOR. The probability of guidewire crossing ≤30 minutes declined consistently from 100% to 0% according to the KCCT score ( P <0.01, all). The KCCT score showed higher discriminative performance compared with the other scoring systems (c-statistics=0.78 versus 0.65–0.72, P <0.001, all). The sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of a KCCT score of <4 for guidewire crossing ≤30 minutes was 70%, 68%, 72%, 73%, and 70%, respectively. The KCCT score also showed consistent results with procedural success ( P <0.05, all). These results could be reproduced in validation data set ( P <0.05, all). Conclusions—: KCCT scoring could predict successful guidewire crossing ≤30 minutes and also procedural success. KCCT scoring may enable noninvasive grading difficulty of CTO percutaneous coronary intervention.

We validated and compared transluminal attenuation gradient (TAG) and corrected coronary opacification (CCO) of coronary computed tomography angiography (CCTA) with invasively measured fractional flow reserve (FFR). One of the major limitations of CCTA is the discrepancy between angiographical stenosis and ischaemia-causing stenosis. Recently two new CCTA analysis methods, TAG and CCO, have been attempted to overcome this limitation but without physiological validation. We measured TAG and CCO of 97 major epicardial coronary arteries from 63 patients who underwent CCTA and followed by invasive coronary angiography and FFR. Diagnostic performance of TAG and CCO was assessed using FFR <0.80 as the reference standard. The overall diagnostic performance of TAG and CCO on a per-vessel basis was moderate and similar (c-statistic = 0.696 vs. 0.637, P = 0.29). The sensitivity, specificity, positive, and negative predictive values of TAG cut-off ≤−0.654 for FFR 0.063 were 65.0, 61.4, 54.2, and 71.4%. TAG showed an incremental value to the diagnostic performance of CCTA but CCO did not (c-statistic =0.726 vs. 0.809, P = 0.025; c-statistic =0.726 vs. 0.784, P = 0.09). In net reclassification improvement (NRI) analysis, addition of TAG to CCTA did not result in significant reclassification (NRI = 1.0%, P = 0.41) and addition of CCO to CCTA resulted in negative reclassification (NRI = −9.3%, P = 0.036). Intracoronary attenuation-based CCTA analyses, TAG and CCO, showed moderate correlation with physiological coronary artery stenosis. The incremental value of TAG or CCO to the evaluation of haemodynamically stenosis by CCTA seemed to be limited.