X-ray vision: Diagnostic accuracy, visual search patterns and 'expert' image interpretation

Diagnostic performance in radiological interpretation increases with experience, though expert-level visual search patterns appear to develop before expert diagnostic accuracy, according to results of a new study published in the journal Radiology.

In the past, expert consultant radiologists have shown the ability to make diagnostic decisions based on image interpretation in as fast as 0.25 seconds, less time than it takes the average person to view an image in its entirety.

How this level of accuracy and efficiency is developed has been the subject of numerous studies, many of which focus on eye-tracking analysis of experts’ visual search patterns, said lead author Brendan Kelly, MD, and his colleagues from Mater Misericordiae University Hospital in Dublin, Ireland.

“Studies have shown significant divergence in the search pattern of experts and naive viewers when looking at both natural and radiologic images,” the authors wrote. “However, to date, reports of the use of both eye tracking and diagnostic accuracy to investigate how the development of radiologic expertise over different phases of hospital doctors’ careers have not been published.”

Kelly and his team set out to investigate this development by measuring both diagnostic accuracy and radiologist eye movement during visual search of chest radiographs for pneumothorax, or collapsed lung.

To do so, they created four clinician groups representing varying levels of professional experience: An expert group of five consulting radiologists; four radiology registrars; five senior house officers; and six interns. Participants were asked to assess 30 chest radiographs, 14 of which had a collapsed lung, and indicate whether they believed a pneumothorax was present. A receiver operating characteristic (ROC) curve analysis was carried out on diagnostic decisions and participants’ eye movements were recorded and analyzed.

They found that diagnostic accuracy improved significantly with experience, with the average area under the ROC curve being 0.947 for consultants, 0.792 for registrars, 0.693 for senior house officers, and 0.659 for interns. Eye-tracking metrics also decreased with experience, with the largest gap in visual search performance present between registrars and senior house officers. Overall reading times also decreased with experience for each of the clinician groups.

“Perhaps most important is the fact that, for this search task, significant differences in diagnostic performance and search pattern (eye-tracking metrics) appeared at different stages of the medical career,” the researchers wrote. “The possibility of a dissociation between the rate of change in visual search and diagnostic accuracy warrants attention.”

While the study focused only on the task of identifying a pneumothorax in chest radiographs, Kelly and his colleagues believe their findings open the door to expanded research into the development of expert-level image interpretation and, potentially, better training for radiology professionals.

“If these findings can be shown to hold true for other search tasks,” the authors concluded, “this may suggest that specific training in areas appropriate to the development of the trainee could improve radiology expertise.”