Evaluation of implementation of artificial intelligence (AI)-assisted echocardiography in clinical practice

Abstract

Abstract Background Transthoracic echocardiography (TTE) is critical for detecting and managing cardiac dysfunction. However, its dependence on expert acquisition means its accessibility in rural areas may be limited, leading to potential missed diagnoses and delayed management. AGILE-Echo is a trial of Artificial intelligence-guided (AI)-TTE for permitting non-expert image acquisition in rural and remote Australia. Aims We aimed to assess markers of image quality to determine 1) Rates of diagnostic image acquisition at sites, 2) Whether a learning curve exists, 3) Whether patient demographic details influence the diagnostic quality of images? Methods 109 (51% M, age 66 ±15 BMI 30±7) unselected participants were recruited through 6 sites for evaluation of exercise intolerance (73%) or heart valve disease. We also assessed whether images were of "diagnostic" quality. Studies were also graded if each study's PLAX, PSAX, and AP4CH were visible. Regression models were used to determine the significance of the patient's clinical background to diagnostic image quality. Results Participants had substantial rates of ex- or current smoking (65%), HTN (57%), and AF (13%), with an overall 4-year ARIC-HF risk of 7.3±11.9%. Clear differences existed between TTE windows and diagnostic quality. The parasternal windows had higher success rates than the apical windows, with the AP2CH having the lowest success rate (Table 1). Increased BMI, HTN, and ARIC-HF score were associated with non-diagnostic images (Table 2). Comparing each assessor’s first 10 studies and eleventh onwards, the percentage of diagnostic images obtained (PLAX 62% A4C 47%) were no different from those with fewer (PLAX 68%, p=0.71; A4C,35%, p=0.21). Conclusions Rates of successful image acquisition by non-experts using AI-guided echo were dependent on the window acquired, with the PLAX being most successfully acquired and patient clinical demographics, including age, BMI, and HF risk, showing an association with image quality. After 10 studies, there was no clear learning curve for improvement in diagnostic quality.

Ähnliche Arbeiten