I see something you do not: Eye movement modelling examples do not improve anomaly detection in interpreting medical images

Abstract

AbstractBackgroundWhen interpreting medical images such as dental panoramic radiographs (Orthopanthomogram, OPT), errors are frequent. Previous research has shown that eye movement modelling examples (EMME) are a supportive training method for medical image interpretation to reduce errors. To date, EMME support for OPTs has not been verified.ObjectivesWe investigated whether a training with EMME and verbal explanations supports dental students in evaluating OPTs.MethodsDental students were randomly assigned to an intervention (N = 42) or a control group (N = 41). The intervention group received the EMME between pre‐ and post‐test. In a laboratory study, we measured students' gaze behaviour during evaluating OPTs and the detection rate of anomalies.Results and conclusionsThe training led to fewer, shorter, and later fixations on anomalies and no difference in visual coverage of the OPT. The detection rate of anomalies did not improve. We replicated the latter finding in an online study (N = 31). Students may not have been able to apply the information from the EMME to detect anomalies. The image reading processes changed to more efficient rather than deeper visual search.Major takeawaysThis study evaluated a training method with EMME for anomaly detection of OPTs. EMMEs did not improve the anomaly detection of dental students and changed their visual search process toward a more efficient rather than a deeper search.