Impact of real-time AI guidance on focused cardiac ultrasound acquisition time in residents: a pilot randomized crossover trial

Background Focused cardiac ultrasound (FoCUS) is widely used in acute care, yet evidence for real-time artificial intelligence (AI) guidance in resident-performed FoCUS is limited. Objectives We conducted a single-center pilot randomized crossover trial to evaluate whether real-time AI guidance reduces residents’ time to acquire diagnostic-quality FoCUS views, improves image acquisition success, and changes self-reported confidence. Methods Participants were recruited at the [remove] Hospital. Each resident performed FoCUS under two conditions with real-time AI guidance (AI+) and without (AI−). Primary outcomes were acquisition time and the success rate of acquiring diagnostic-quality images for the five basic FoCUS views. Secondary outcomes were acquisition time and success rate across a total of 10 views, consisting of the basic views plus five additional advanced FoCUS views. We also evaluated changes in self-reported confidence using a questionnaire. Results Nineteen residents were included in the analysis. For the five views, total acquisition time did not differ significantly between AI+ and AI− (192.9 ± 75.8 s vs. 125.0 ± 41.7 s; mean difference, −67.9 s [95% CI, −138.7 to 2.9]; P = 0.058). The success rate of acquisition showed no significant differences. For the ten views, AI guidance increased success rate for apical two- and five-chamber views (37% to 84%; P = 0.016, 47% to 84%; P = 0.023). Self-reported confidence improved overall, but not for four- and five-chamber views. Conclusions Real-time AI guidance did not significantly reduce the time to acquire basic FoCUS views, nor did it significantly improve the success rate.