Expert-Level Agreement and Intermodel Convergence of Contemporary AI Chatbots in Confronting Common Radiology Misconceptions: A Cross-Sectional Study

Abstract

Misconceptions about diagnostic imaging modalities, including magnetic resonance imaging (MRI), computed tomography (CT), ultrasound, and positron emission tomography (PET), remain common and may contribute to patient anxiety, delayed investigations, and increased clinician workload. With growing reliance on artificial intelligence (AI) chatbots for health information, evaluating the reliability of contemporary publicly accessible models is increasingly important.

In this cross-sectional validation study, 58 statements representing common radiology myths and true clarifications across MRI, CT, ultrasound, and PET were compiled from authoritative sources. Each statement was queried using standardized prompts across three freely available AI chatbots: ChatGPT (GPT-5.2), Google Gemini (Gemini 3 Flash), and DeepSeek (DeepSeek-V3.2). Responses were classified as “myth” or “fact” and compared with expert consensus.

All three chatbots demonstrated complete agreement with expert consensus across all 58 statements. No false classifications or intermodel discrepancies were found, and explanations aligned with established radiological principles.

Contemporary AI chatbots demonstrate high reliability and convergence in addressing common, low-ambiguity radiology misconceptions, reflecting stability of foundational imaging knowledge; however, findings are domain-specific and require cautious extrapolation with ongoing evaluation.