Real world evaluation of artificial intelligence echocardiography image guidance and acquisition with novice scanners in multiple clinical settings

Abstract

Abstract Funding Acknowledgements Type of funding sources: Private company. Main funding source(s): Caption Health, US OnBehalf Beacon Hospital Research Institute Background Cardiac echocardiographic scanning requires significant training and experience. The FDA approved Caption Echocardiography system uses a deep learning artificial intelligence software that guides novice scanners to optimal position and then automatically acquires the highest quality image. Further advances to this novel technology have allowed for immediate calculation of left ventricular ejection fraction directly from these images. Methods Following only brief training, we sought to evaluate study quality by novice scanners in four clinical settings, both acute and ambulatory – the Emergency Department, Intensive Care Unit, Heart Failure clinic, and the Oncology Day Unit. 120 patients (30 per clinical area) were recruited and underwent 2 echo scans – one by a novice scanner using the Caption AI system, and one by an expert scanner using the same ultrasound system but without AI guidance. Both studies were evaluated blindly and independently side by side by 3 accredited experts judging diagnostic quality. ‘AutoEF’ measurements were compared with expert scanners measurements by Simpson’s biplane technique. Results 120 patients with mean age 62.04 years old (range 18 – 92) were enrolled, consisting of 30 from each clinical area. 8 novice scanners – nursing staff and junior doctors – scanned 15 patients each, allowing their first 3 cases to be excluded. 75% of patients were male. BMI ranged from 18 – 37 with mean 27.48. Image quality allowed left ventricular function to be sufficiently captured in 97.5% of studies, right ventricular function in 95% and outrule of pericardial effusion in 97.5%. ‘AutoEF’ was determined in 88.3% of studies and matched the expert measurement in 84% of cases. Blinded expert assessment found that 60% of Caption studies were of equal or better quality than the experienced sonographer scan. Conclusion The Caption AI technology safely allows novice users to provide efficient and accurate point of care echo in differing clinical settings to a standard comparable to expert scanners, and automatically determines left ventricular ejection fraction.