Medical-AI trial: when should AI advice appear? eye-tracking of visual search, diagnosis, and trust in chest X-ray reading (human-AI interaction) (NCT07675694)
A within-subject interventional study using eye-tracking to examine whether showing AI advice before or after a clinician first reviews a chest X-ray changes visual search, reading time, diagnostic decisions, confidence, and trust in AI. Healthcare professionals complete two reading sessions. 24 participants.
Trial overview (primary data)
- StatusRecruiting
- ConditionsDiagnostic Imaging, Eye Tracking, Artificial Intelligence (AI)
- InterventionsBEHAVIORAL: Original CXR First Timing, BEHAVIORAL: AI Output First Timing
- SponsorUniversity Hospitals, Leicester
- Target enrollment24 participants
- Period2026-06-30 〜 2026-12-01
Key points
- A within-subject interventional study using eye-tracking to see how showing AI advice before vs. after affects visual search, time, decisions, confidence, and trust in AI.
- The same person experiences both conditions; participants are healthcare professionals completing two chest X-ray reading sessions.
- The problem: it is unclear whether AI-information timing changes how images are viewed, decisions are made, and AI is used.
- Early presentation risks anchoring/automation bias; later presentation preserves independent judgment — the study captures this interaction with objective data.
- 24 participants, small and exploratory; it does not establish an optimal timing or clinical superiority.
Chest X-rays are widely used to diagnose lung and heart conditions, and AI is increasingly joining in to support their reading. An easily overlooked question is when to show the AI advice. If the AI result is seen before reading, the clinician gaze and judgment may be pulled toward that answer (so-called anchoring or automation bias).
Conversely, reading first and seeing AI afterward may better preserve independent judgment but make it harder to take in what the AI noticed.
This study tackles that timing effect directly with eye-tracking (a technology that measures where and how long someone looks).
Per the registry summary, it sets up a condition where the AI support information is shown before the healthcare professional first views the image and one where it is shown after, and, with a within-subject design in which the same person experiences both, compares how visual search, time spent reading, diagnostic decisions, confidence, and trust in AI change.
Participants are healthcare professionals who complete two chest X-ray reading sessions. It is a small study of 24 participants.
Debate about medical AI tends to center on how accurate the AI is, but in real practice how people receive and use AI shapes outcomes. A seemingly minor design choice, the order in which advice is shown, can change where attention goes, trust, and the final decision. Capturing this human-AI interaction with objective data (gaze) is meaningful as a design discipline for using AI safely.
That said, this is a small, exploratory study; it does not establish the optimal timing or clinical superiority.
Why it matters
The success of medical AI depends not only on AI accuracy but on how people receive and use it. Capturing with objective data how the order of showing advice changes attention, trust, and decisions is a useful reference for the design discipline of using AI safely (this trial is small and exploratory, not an establishment of optimal timing).
FAQ
What does eye-tracking reveal?
Did it show that seeing AI first is good or bad?
Sources (primary)
Source: ClinicalTrials.gov (U.S. NIH/NLM, public domain). This site does not provide medical advice. Verify the latest and exact details with the official source. This site is not endorsed or certified by the NIH/NLM.
- ClinicalTrials.gov (study record, original)
- NCT ID: NCT07675694