Visual search behaviour in skeletal radiographs: a cross-specialty study

Area of Science:

• Radiology
• Orthopaedics
• Medical Imaging Analysis
• Human Factors in Medicine

Background:

• Accurate fracture identification in plain radiographs is crucial for patient management.
• Visual search strategies can vary significantly among medical professionals with different experience levels.
• Understanding these differences can optimize training and improve diagnostic accuracy.

Purpose of the Study:

• To evaluate if experience enhances the consistency of visual search behavior in identifying fractures on radiographs.
• To determine the impact of medical specialization on visual search patterns during radiographic interpretation.
• To correlate visual search phases (discovery and reflective) with diagnostic confidence.

Main Methods:

• Twenty-five observers (radiologists, orthopaedic surgeons, registrars, house officers) analyzed 33 skeletal radiographs (shoulder, hand, knee).
• Eye-tracking data (Tobii 1750) and diagnostic confidence levels were collected simultaneously.
• Kullback-Leibler (KL) divergence and Gaussian mixture models quantified search consistency and phase relationships.

Main Results:

• Expert groups showed higher true positive rates (p<0.001) and reduced dwell time on fracture sites (p<0.001).
• Experts exhibited smaller KL distances (p<0.001) between trials, indicating greater consistency.
• The reflective search phase duration decreased with increased diagnostic confidence (r=0.266, p=0.074).

Conclusions:

• Expert visual search behavior is characterized by higher accuracy and consistency, with efficient fixation on fracture sites.
• This expert strategy aligns with the global-focal model, suggesting reflective search aids in cross-referencing and decision-making.
• Specialization appears to be a stronger driver of altered search strategy than the duration of training.