A within subject diagnostic behaviour study using eye tracking to evaluate how the timing of AI results, as presented either before or after the primary chest X-ray, affects visual search behaviour, diagnostic accuracy, and trust in clinicians interpreting chest radiographs

To note, data hasn’t been collected or analysed, that this abstract sets out the framework for this research study.

Artificial intelligence (AI) tools are increasingly integrated into clinical imaging to support interpretation, yet little empirical evidence exists on how clinicians interact with them during real diagnostic tasks. The study investigates how the timing of AI result presentation, whether shown before (AI first) or after (AI second) the clinician’s initial review of a primary chest X-ray (CXR), affects diagnostic behaviour, visual search, and trust.

Using a within subject, counterbalanced design, clinicians complete two eye tracked reporting sessions under both timing conditions, evaluating the same anonymised CXR drawn from routine NHS practice. Each image is categorised by verified ground truth and AI outcome class (true/false positive or negative). Automated wearable eye tracking glasses and structured reporting forms capture measures of visual attention, diagnostic accuracy, interpretation time, and self-reported trust.

Primary analyses test whether AI timing alters search behaviour, accuracy, or efficiency. Secondary and exploratory analyses assess confidence, reliance, and subgroup differences between expert and non-expert readers, case difficulty and AI blind spots (false negatives/positives). A complementary mixed methods survey of clinicians, patients, and the public explores wider perceptions of trust, responsibility, and acceptability of AI in imaging.

This research will generate new empirical evidence on how AI CXR output timing influences diagnostic performance, visual search practice and human AI interaction in radiology. That this can be used to inform safe, trustworthy design and management of AI software systems intended specifically for evaluating CXR in clinical NHS practice.

Outputs

None so far, though I am planning to produce proffered papers, as well presentations and posters in appropriate professional journals and meetings. Here are some of the journals and professional bodies I will seek to disseminate my research works (this is suggested and not confirmed).

Journals

• Clinical Radiology (Royal College of Radiologists)
• Radiography (Society of Radiographers)
• British Journal of Radiology (BJR)
• Insights into Imaging and Artificial Intelligence in Medicine

Professional organisations for presentation at meetings

• RCR Annual AI Congress and RCR Annual Conference
• UK Imaging & Oncology Congress (UKIO) – joint RCR, SoR, and IPEM event
• Society of Radiographers Annual Radiography Conference
• IPEM Medical Physics and Engineering Conference
• European Congress of Radiology (ECR)