Radiographer AI study_Automation Bias and Decision Switching Datasets

AI is becoming more prevalent in healthcare across the world. As staff shortages and increased use of radiology services continue, AI has been proposed to support the heath service. The interaction of human end-users with AI is currently not well understood. This is of paramount importance when considering a future where AI will be used in patients’ care pathways.
This study investigated the impact of different forms of AI feedback on student and qualified radiographers’ diagnostic accuracy and likelihood to change their mind from their initial diagnosis. Participants were recruited from around the world and presented with different types of AI feedback (heatmaps and binary diagnosis). This study found that:
• AI feedback improves radiographers’ and student radiographers’ diagnostic accuracy on plain radiographic images, except when the AI feedback was inaccurate and in pathological cases in the student group.
• Heatmaps reduced diagnostic accuracy, while textual (binary) diagnosis had a positive impact.
• Decision switching was more prevalent in the student group
• Automation Bias was present in both student and radiographer groups but had greater prevalence in the student group.

This dataset of clinical images is restricted as they are real patient images. Access to the data may be applied for following instructions provided here

These data support the publication entitled ‘The impact of AI feedback on the accuracy of diagnosis, decision switching and trust in radiography.’

There are multiple datasets:
1) Restricted clinical images
2) Original SPSS dataset of responses to AI presentation
3) Original Excel dataset of responses (cleansed for SPSS)

This work has been funded by the College of Radiographers Research Industry Partnership Research awards scheme (CoRIPS) no. 183.