P21 Artificial intelligence assessment of the thoracic aorta is accurate, reliable and has potential clinical impact

• Published in: Heart (British Cardiac Society) Vol. 108; no. Suppl 2; p. A10
• Main Authors: Harris, Maredudd, Graby, John, Jones, Calum, Waring, Harry, Lyen, Stephen, Hudson, Ben, Rodrigues, Jonathan
• Format: Journal Article
• Language: English
• Published: London BMJ Publishing Group Ltd and British Cardiovascular Society 21.09.2022; BMJ Publishing Group LTD
• Subjects: Accuracy; Aneurysms; Artificial intelligence; Coronary vessels; Scientific poster presentations
• ISSN: 1355-6037; 1468-201X
• DOI: 10.1136/heartjnl-2022-BSCI.26

ObjectiveTo assess the diagnostic accuracy, reliability and clinical impact of artificial intelligence (AI) derived thoracic aorta analysis (AI-Rad Companion, Siemens) on routine clinical gated and non-gated chest CT.MethodsThis was a single centre retrospective study. AI diagnostic accuracy was assessed on 210 consecutive CT aortas and compared to cardiothoracic radiologist reference standard. AI test-retest accuracy was assessed on immediate sequential pre- and post-contrast CT aortas in 29 patients. Real-world AI clinical impact was assessed in 197 non-gated CT chests with comparison to manual radiology reports and patient electronic records to establish the detection rate of previously unknown aortopathy.ResultsAI analysis was feasible in 97% (421/436 scans). Diagnostic accuracy of AI was good to excellent (intraclass correlation coefficient [ICC] 0.87–0.96). Test-retest accuracy of expert reader (ICC 0.88–0.98) and AI (ICC 0.82–0.94) for the ascending aorta were good to excellent. AI identified new aortopathy in 27% of non-gated scans versus routine clinical reports (X2 51, p<0.001).ConclusionAI provides measurements of the thoracic aorta comparable to an expert reader with similar reliability. Whereas manual reporting of non-dedicated studies significantly underreports thoracic aneurysms, AI identifies previously unknown aortopathy in a significant proportion (27%) of non-gated CT chests. The use of AI software in non-dedicated CT chest imaging could support earlier diagnosis of thoracic aneurysms before potentially fatal complications.