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

Aim

To determine whether experience improves the consistency of visual search behaviour in fracture identification in plain radiographs, and the effect of specialization.

Material and methods

Twenty-five observers consisting of consultant radiologists, consultant orthopaedic surgeons, orthopaedic specialist registrars, orthopaedic senior house officers, and accident and emergency senior house officers examined 33 skeletal radiographs (shoulder, hand, and knee). Eye movement data were collected using a Tobii 1750 eye tracker with levels of diagnostic confidence collected simultaneously. Kullback–Leibler (KL) divergence and Gaussian mixture model fitting of fixation distance-to-fracture were used to calculate the consistency and the relationship between discovery and reflective visual search phases among different observer groups.

Results

Total time spent studying the radiograph was not significantly different between the groups. However, the expert groups had a higher number of true positives (p<0.001) with less dwell time on the fracture site (p<0.001) and smaller KL distance (r=0.062, p<0.001) between trials. The Gaussian mixture model revealed smaller mean squared error in the expert groups in hand radiographs (r=0.162, p=0.07); however, the reverse was true in shoulder radiographs (r=−0.287, p<0.001). The relative duration of the reflective phase decreases as the confidence level increased (r=0.266, p=0.074).

Conclusions

Expert search behaviour exhibited higher accuracy and consistency whilst using less time fixating on fracture sites. This strategy conforms to the discovery and reflective phases of the global–focal model, where the reflective search may be implicated in the cross-referencing and conspicuity of the target, as well as the level of decision-making process involved. The effect of specialization appears to change the search strategy more than the effect of the length of training.