Measuring the arterial phase of the right coronary artery in the patients suspected of coronary artery disease: a dual study by dynamic angiography and deep learning program

Abstract

Abstract Funding Acknowledgements Type of funding sources: None. Background. In the diagnosis of coronary artery disease (CAD), coronary angiography (CA) plays a crucial role in determining the location and severity of the stenosis, the anatomical aspect of a lesion. It does not accurately reflect the flow dynamics in the coronary artery. This study aimed to evaluate the coronary flow abnormalities based on our new angiographic technique and Deep Learning (DL) program in patients suspected of CAD. Methods. We randomly selected patients who were admitted with suspected CAD. All patients underwent our new technique of CA. After the index coronary artery was filled completely with contrast, we stopped the injection. At that time, the blood in white color flew in. The flow characteristics, the shape of the tip, borders, and direction could be clearly observed above a black background of the contrast. In this study, we measured the arterial phase (AP) from the beginning when the blood moved in until the end when all contrasts in black color washed out of the distal vasculature. In the DL protocol, the U-Net model combined with Dense-Net-121 and a binary image classification model are used to predict the beginning and ending frame. To obtain the best image for the DL program, we analyzed only the flow of the right coronary artery (RCA). Results. 81 patients were enrolled. In patients with normal coronary angiography, the mean AP was 1.86s (27.4 +/- 5.4 frames). In patients with one significant lesion, the mean AP value was 2.35s (35.3 +/- 7.7 frames). The mean difference of the AP between the two groups was 0.49s (95% confidence interval: 0.295 to 0.694). This difference is statistically significant. Our DL has the mean root square error in predicting the AP was 0.34s. Conclusion. In patients with CAD, the prolonged arterial phase could be accurately estimated using the DL program, reflecting the slow circulation of highly oxygenated blood. It could be used as a marker of coronary perfusion in future studies.