ABSTRACTS
AI-Driven Ultrasound Training for Prehospital Personnel: A Study of Diagnostic Improvements in Cuenca, EcuadorAuthor: Fernanda Pacheco, Ariel Quezada, Maria G. Icaza, Maria Peralta, Rosalia Aray, and Mateo Godoy | | Associate Authors:
Introduction: New ultrasound (US) devices are smaller and portable, making them perfect for use in the prehospital setting. However, this imaging modality is heavily dependent on training and prone to user error without it. We created a hands-on US training module integrating an artificial intelligence (AI)–based teaching method to improve the skills of prehospital providers in diagnosing thoracic pathology. Objective: To evaluate US diagnosis following completion of an AI-based US training module. Methods: This prospective observational cohort study was conducted across two prehospital management institutions in Cuenca, Ecuador, encompassing 24 participants (12 from each institution). Initially, participants were assessed using a dichotomous questionnaire featuring thoracic images to distinguish between pathological and normal findings. Group 1 (12 participants) received a conventional 2.5-hour hands-on US training symposium, whereas group 2 (12 participants) received identical training supplemented with AI assistance through automatic localization and recognition of anatomical structures. Statistical analysis was performed using a two-sample t-test to compare mean differences between the groups. Results: Exclusion criteria included prior US experience, resulting in a final sample of 24 participants. The variables analyzed were pre- and postsymposium scores and the influence of AI on diagnostic accuracy for thoracic pathologies. In group 1 (non AI), the pretest mean score was 10.17 (SD = 2.125) and the posttest mean score was 10.5 (SD = 1.314). In group 2 (with AI), the pretest mean score was 11.25 (SD = 1.658) and the posttest mean was 14.50 (SD 1.0). The paired samples test for group 1 showed a mean difference of −0.333 (SD = 2.839) with a 95% confidence interval (CI) of −2.137 to 1.471 and a p-value of .346, indicating no statistical significance. Conversely, group 2 demonstrated a mean difference of −3.250 (SD = 0.524) with a 95% CI of −4.403 to −2.097 and a p-value of 0.000, indicating a statistically significant improvement. Conclusion: Our AI-based teaching method helps improve US training for prehospital providers, specifically in identifying emergent thoracic pathology. It reveals how broadly this tool can be used for teaching in environments with little access to US skills. Critically, this skill can be applied to lower-resource prehospital environments and impact clinical decision-making.
|
