The Impact of Coronary CT Angiography Image Quality on Artificial Intelligence Technology's Ability To Automatically Identify Coronary Artery Lesions And Stenosis

Abstract Background To explore the impact of image quality on the ability of artificial intelligence (AI) to identify coronary artery lesions and detect the degree of coronary artery stenosis in coronary artery CT angiography (CCTA), using invasive coronary angiography (ICA) as the reference standard. Methods 156 patients suspected of coronary heart disease were enrolled from June 2020 to June 2021, including 92 males and 64 females, with an average age of 66.1 ± 9.4 years. Left anterior descending (LAD), right coronary artery (RCA), and left circumflex (LCX) were divided into non-stenosis, mild stenosis, moderate stenosis, severe stenosis, and occlusion according to ICA as the gold standard. With image quality as the categorical variable, each vessel was evaluated to investigate the effect of image quality on the performance of AI technology in automatically detecting coronary artery lesions and the degree of CCTA. Results Compared with traditional manual postprocessing (13.9 ± 3.6min per case), the average postprocessing and interpretation times of AI were 5.4 ± 1.5min per case were significantly reduced; there was no significant difference in the detection of plaques between the coronary AI technology and ICA (P = 0.213). The consistency of AI in assessing the stenosis degree of 156 subjects were medium consistent with that of ICA, with P value < 0.001. With image quality as the categorical variable, the accuracy of human + AI in evaluating the degree of LAD, RCA stenosis was significantly higher than that of AI (P < 0.001), with OR values of 5.258 (2.451,11.376) and 6.552 (2.989,14.362) for the image quality of medium, with OR values of 8.200 (2.747,24.478) and 14.333 (3.886,52.874) for image quality of high, respectively; and the accuracy of human + AI in evaluating the degree of LCX stenosis was significantly higher than that of AI (P < 0.05), with OR values of 2.179 (1.063,4.468) and 3.266 (1.198,8.903) for image quality of high and medium, respectively. Conclusion Compared with traditional CCTA manual processing, coronary artery AI-assisted diagnosis technology has a shorter processing time, consistent plaque detection with ICA, and the ability to evaluate the degree of stenosis of each coronary artery under the supervision of a physician, thereby improving work efficiency.