Clinicians' Trust in AI-Based Fetal Growth Estimates

Overview

This study examines how clinicians trust and use artificial intelligence (AI) when estimating fetal weight during pregnancy. Accurate assessment of fetal growth is important for identifying growth problems that may affect pregnancy management. New AI-based tools can estimate fetal weight from ultrasound images, but little is known about how clinicians trust these estimates or how uncertainty information influences their decisions. In this study, clinicians will review anonymized ultrasound cases and compare fetal weight estimates generated by an AI model with traditional estimates. Some clinicians will also be shown information about the AI model's performance and uncertainty, while others will not. Participants will be asked to choose which estimate they find most reliable, indicate their level of confidence, and decide whether they would recommend follow-up scans. The study aims to better understand how AI and uncertainty information affect clinical decision-making and trust among clinicians with different levels of experience.

This is a randomized, matched, vignette-based questionnaire study designed to investigate clinicians' trust in and use of AI-based fetal growth estimates. Clinicians from obstetrics and gynecology departments will be recruited and stratified by experience level. Participants will be randomized to either a control group or an intervention group. The intervention group will receive brief information about the AI model's overall performance, while the control group will not receive this information. Each participant will assess a set of anonymized third-trimester ultrasound cases. For each case, clinicians will be presented with standard ultrasound images and relevant clinical context. They will be shown fetal weight estimates generated by an AI-based model and by a traditional biometric method, with or without accompanying uncertainty information in the form of confidence intervals. For each case, clinicians will select the estimate they consider most clinically reliable, rate their confidence in that choice, and indicate whether they would recommend a follow-up growth scan. Case sets are matched by clinical experience, ensuring that identical cases are evaluated by clinicians with similar backgrounds across study arms. The study focuses on clinicians as participants and involves no patient intervention. All ultrasound data are fully anonymized. The results will provide insight into how AI-generated estimates and uncertainty information influence clinical trust, preferences, and decision-making in fetal growth assessment.