Task-shifting to nonexperts using artificial intelligence-guided point-of-care ultrasound: a cohort study of patient selection, image quality, and learning curves.
Authors
Affiliations (9)
Affiliations (9)
- Imaging Research Laboratory, Baker Heart and Diabetes Institute, Melbourne, Victoria 3004, Australia.
- Baker Department of Cardiometabolic Health, University of Melbourne, Melbourne, Victoria 3004, Australia.
- Ochre Medical Centre, Huonville, Tasmania, Australia.
- Alice Springs Hospital, The Gap, Northern Territory, Australia.
- Western Australian Country Health Service, Merredin District Hospital, Wheatbelt, Western Australia, Australia.
- Walgett Aboriginal Medical Service Limited, Walgett, New South Wales, Australia.
- Nepean Clinical School, University of Sydney, Sydney, New South Wales, Australia.
- Princess Alexandra Hospital, University of Queensland, Woolloongabba, Queensland, Australia.
- Menzies Institute for Medical Research, Hobart, Tasmania, Australia.
Abstract
To define rates of diagnostic image acquisition, clinical drivers of image quality and the learning curve for artificial intelligence (AI)-guided image acquisition on point-of-care ultrasound (AI-POCUS) in rural and remote communities. AI-guided image acquisition on point-of-care ultrasound was performed using AI software integrated with a desktop ultrasound system in 181 participants (65 ± 15 years, 47% female). A standardized training protocol included online material, lab attendance for 1 day, and online mentoring. Diagnostic-quality images were obtained from 72% of parasternal and 55% of apical images (<i>P</i> < 0.001). Scans were classified as 'Diagnostic' if diagnostic-quality images were obtained in the majority of parasternal and apical views (ACEP score ≥3) in ≥50% of windows for both apical and parasternal views. Body surface area (BSA) [OR 0.16 (0.05;0.50), <i>P</i> = 0.002] and hypertension [OR 0.50 (0.27;0.93), <i>P</i> = 0.03] were associated with diagnostic image quality in the apical window, whereas only hypertension [OR 0.43 (0.20;0.88), <i>P</i> = 0.024] was associated with diagnostic quality in the parasternal window. The learning curve was assessed by comparing the quality according to quantity of scans performed and professional background (nurse, health worker, or general physician). Physician-acquired scans [OR 3.85 (1.92;8.33), <i>P</i> < 0.001], scan 11th onwards [OR 2.86 (1.45;5.56), <i>P</i> = 0.002], and users who performed ≥20 scans [OR 3.58 (1.79;7.14), <i>P</i> < 0.001] predicted study completeness. In rural community practice, the learning curve associated with AI-POCUS diagnostic quality seems longer than reported in other studies from inpatient settings. In novice users, diagnostic quality is greater in the parasternal than the apical windows.