Identification of the low wall shear stress area in the middle and distal coronary artery: a dynamic angiography and deep learning analysis

Abstract

Abstract Background Based on the principle and practice of fluid mechanic, when applying to the coronary circulation, at the location of turbulence (collision of the antegrade against the retrograde flow), stagnant flow was of low velocity and low wall shear stress (WSS). This low WSS triggers the atherosclerotic process. Furthermore, our prior studies showed that turbulent flow inflicted the first damage to the intima in young people, even before the effects of hypertension or hyperlipidemia kick in. Turbulence also promoted faster growth in the plaque by cracking the cap of vulnerable plaque which was translated into acute coronary syndrome. How could we image the flow in low WSS area? How does the stagnant and turbulent flow effect on the wall of coronary artery? In this study, could dynamic angiography and deep learning model aswer the above questions. Method Patients underwent a new dynamic angiography and were sequentially selected if they had one moderate lesion. At first, after the coronary arteries were filled completely with contrast (in black color), the blood (in white color) was seen moving in and recorded at 15 frames per second. All characteristics of the blood flow, its direction, movement, shape of the tip and velocity were identified and tabulated. At the same time, an artificial intelligence program was designed based on Deep Learning Model to analyse the blood flow. First, the segmentation model was built with U-Net and Densenet-121 to recognize the coronary artery, then the Convolutional Neural Network was build to classify the boundary layer, identify the low WSS area and calculate the duration of the arterial phase (AP). Results Thirty-two patients (23 males) with forty-five lesions were enrolled and their data analyzed with Quantitative Coronary Analysis (QCA) and Deep learning (Figure 1). In the group of patients with stagnant flow (low velocity – low WSS), the mean time of contrast hangover (arterial phase) at the internal-boudary layer was 2.61 +/- 0,66 millisecond. In contrast, in the patients with no stagnant flow, the mean time was 1.86 +/- 0.41 millisecond, statistically significant with p =0.02 (Fig 2). Furthermore, the group with stagnant flow has thirty-one lesions in comparison with fourteen the group with no stagnant flow. Conclusion The turbulent and stagnant flow were successfully detected and measured by the new dynamic angiography and Machine Learning program. The prolonged contrast stagnation at the boundary layer identified the area with low velocity and low WSS which was the culprit in triggering the atherosclerotic process. This stagnant flow also delayed the delivery of new blood to the distal myocardium and could cause chest pain. It is important to understand and image the mechanism of CAD in order to further the understanding and management of this condition.Deep learning model suggest the stenosisArterial Phase Time of Coronary Artery