Simulated tumor recognition in mammograms is biased by serial dependence

Abstract

In radiological screening, radiologists scan myriads of radiographs with the intent of recognizing and differentiating cancerous masses. Even though they are trained experts, radiologists’ human search engines are not perfect: average daily error rates are estimated around 3-5%. A main underlying assumption in radiological screening is that visual search on a current radiograph occurs independently on previously seen radiographs. However, recent studies have shown that our current perception is biased by previously seen stimuli (Fisher & Whitney, 2014; Liberman et al., 2014); the bias in our visual system to misperceive current stimuli towards previous stimuli is called serial dependence. Here, we tested whether serial dependence impacts recognition of tumor-like shapes embedded in actual radiographs. In order to simulate tumors, we created a morph stimulus set based on three canonical shapes (147 simulated tumors in total). On each trial, radiologists were presented with a mammogram containing a random simulated tumor in a random location. Observers were then asked to match the tumor shape they saw by picking the shape from a morph continuum. Twelve radiologists completed 250 trials each. We found that serial dependence affected observers’ recognition of simulated tumors; simulated tumors were perceived as biased towards the simulated tumors seen in previous radiographs for 1 and 2 trials back. Hence, radiological screening was biased towards radiographs presented up to 5-10 seconds ago. Furthermore, regression analysis showed that perception on an average trial was pulled 19% and 6% toward the 1-back and 2-back trials respectively. Taken together, these results suggest that some of the diagnostic errors exhibited by radiologists may be caused by serial dependence from previously seen radiographs.