AI output complexity on chest radiographs alters reading speed but not accuracy in a reader study

In a retrospective reader study of 80 COVID-19 chest X-rays across 8 radiologists, baseline accuracy was high and stayed consistent regardless of AI display design: fewer than 1% of correct calls flipped to incorrect with any AI overlay, while ~1% improved. Complex heatmap+graph…

• Design: Eight readers interpreted 80 COVID-19 CXRs under five conditions (no AI, one-word summary, graph, heatmap, heatmap+graph); eye-tracking and mixed-model accuracy analysis performed.
• Key secondary: Complex AI displays prolonged reading time and increased gaze shifts between image and AI output, signaling higher cognitive load (exact time differences not reported).
• Limitation: Single-center retrospective design with n=80 cases and 8 readers; only COVID-19 detection tested, limiting generalizability to other findings and modalities.

RadPigeon summaries are original and for information only. They are not clinical advice.