Cognitively Biased Prompt Effects on Large Language Model Accuracy for Radiology Board–style Examination Questions

2026/04/15 by Nicholas Dietrich, Dhruv Patel, Joseph Bellissimo +2 · 1 voice · 1 citation
Medicine · #Artificial Intelligence in Healthcare and Education #Radiology practices and education #Clinical Reasoning and Diagnostic Skills

paper · doi:10.1148/ryai.250585

openalex publication_date 2026/04/15 · openalex created_date 2026/04/16 · openalex updated_date 2026/07/31

Abstract

Large language model accuracy for radiology board–style questions was reduced by prompts that mimicked cognitive biases, especially on image-based tasks, and prompt-level mitigations, although effective, only partially restored accuracy.

Cited by

• When Framing Shapes the Answer: Cognitive Bias and Large Language Model Reliability in Radiology

Discussions

• Large language model accuracy for radiology board-style questions was reduced by prompts that mimicked cognitive biases https://doi.org/10.1148/ryai.250585 #OHBM2026 #NeuroRad #brain #radiology [bsky, 4 points, 0 comments]

Related

• Addressing cognitive bias in medical language models
• Large Language Models with Vision on Diagnostic Radiology Board Exam Style Questions
• From radiology findings to artificial intelligence-powered impressions: A retrospective study on the comparative performance of recent large language models
• Evaluating the performance and fragility of large language models on the self-assessment for neurological surgeons
• Investigating the Effects of Cognitive Biases in Prompts on Large Language Model Outputs