The potential of artificial intelligence in the assessment of coronary artery stenosis in follow-up CT examinations

Abstract Background Coronary CT angiography (CCTA) has become one of the main modalities for detecting coronary artery disease (CAD) and is now the first diagnostic choice in many patients. As a result, an increasing number of patients are being identified at an early stage of CAD, allowing effective early medical therapy to be initiated. In addition, CCTA offers the possibility of assessing disease progression and the effectiveness of therapy through follow-up examinations. However, the comparison of different CCTA studies is time-consuming and reader-dependent. The aim of this study was to evaluate the potential of artificial intelligence (AI) for assessing CAD progression in follow-up CCTA examinations. Methods 41 (30 men, 11 women, age: 62.8±7.2 years) patients who were referred for follow-up examinations by their attending physicians were included in this study. Image acquisition was performed on a third–generation dual-source CT scanner using an institutional standard protocol. A series with good image quality was chosen from the CCTA study, transferred to the on-site AI prototype and processed fully automatically. A consensus reading by two experienced readers was used as a reference. Coronary arteries with a diameter 2 mm were excluded from the analysis. Results The median baseline Coronary Artery Disease Reporting and Data System (CAD-RADS) category was 3 (2.0–3.0). The general agreement between AI and reference readers was good (ICC 0.88 (95% CI 0.82-0.92); n=82), with a maximum CAD-RADS difference of 1. After a follow-up period of 2.2±0.7 years, the CAD-RADS category was stable in 36 patients and increased in 5 patients. The agreement between the AI and reference readers regarding the assessment of disease progression was moderate (ICC 0.71 (95% CI 0.45-0.84)), with a maximum category difference of 1. Conclusion AI has great potential for assessing CAD progression in follow-up CCTA examinations. Further developments such as an improvement in artefact detection are required before it can be used in routine clinical practice. AI generated unfolded view