Related Experiment Video
Updated: Jun 30, 2026
05:54
Eye-tracking to Distinguish Comprehension-based and Oculomotor-based Regressive Eye Movements During Reading
Published on: October 18, 2018
Where Your Eyes Go: How AI Output Design Impacts Reading Behavior
Elizabeth A Krupinski1, Marly van Assen2, Carlo N De Cecco2
1Department of Radiology & Imaging Sciences Emory University, 1364 Clifton Rd NE, Atlanta, GA, 30322, USA. ekrupin@emory.edu.
Journal of Imaging Informatics in Medicine
|June 29, 2026
Summary
The design of artificial intelligence (AI) output for chest X-rays impacts radiologist performance. Simpler AI displays may improve efficiency, while complex ones increase interpretation time but can enhance accuracy.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Radiologists interpret chest X-rays for diagnosing conditions like COVID-19.
- Artificial intelligence (AI) tools are increasingly used to aid diagnostic accuracy.
Purpose of the Study:
- To evaluate how different artificial intelligence (AI) output designs influence radiologists' performance in interpreting chest X-rays.
- To assess the impact of AI feedback (none, summary, graph, heatmap, heatmap+graph) on diagnostic accuracy and efficiency.
Main Methods:
- Retrospective study involving eight readers interpreting 80 chest X-rays under five AI conditions.
- Analysis of reader accuracy and eye-tracking data using generalized mixed models and analysis of variance.
- Comparison of diagnostic performance and visual search patterns across varying AI output designs.
Main Results:
- Baseline accuracy for COVID-19 detection was high and minimally affected by AI, with <1% of decisions changing incorrectly.
- Complex AI displays (heatmap+graph) correlated with longer interpretation times and increased gaze shifts.
- Accuracy improvements were observed, with approximately 1% of decisions being corrected by AI.
Conclusions:
- Well-designed AI output can enhance diagnosis accuracy and visual search efficiency in chest imaging.
- Simpler AI displays may facilitate faster decision-making, while complex visualizations can increase cognitive load.
- Optimizing AI presentation is crucial for effective human-AI collaboration in clinical imaging, ensuring the human remains the final decision-maker.
