Abstract
Previous studies of visual search have shown that prevalence rate has a strong impact on an observer's decisional criterion for detecting a target while having minimal impact on d-prime measures of sensitivity (Wolfe & Van Wert, 2010; Evans et al., 2013). Here, we investigated the effect of prevalence rate on a challenging task of lung nodule detection in initially naïve undergraduate participants. Based on our previous work that showed pronounced improvements in lung nodule detection after multi-day training with simulated nodules in 2D chest radiographs (Tong et al., VSS, 2019), we hypothesized that nodule detection training at higher prevalence rates should lead to greater gains in performance, as observers will have had the opportunity to learn from a greater number of positive training examples. In this study, we evaluated the impact of extended training (600 trials over 3 sessions) at lung nodule detection using prevalence rates of 5%, 20% and 50%. Data gathered from 48 participants (16 per group) indicated that low prevalence rates led not only to lower hit rates but also fewer false alarms, with only small differences observed between groups in terms of d’ performance across training sessions. Pre-test vs. post-test measures on a nodule localization task indicated greater improvement in performance after higher prevalence training, with greater gains in performance observed for simulated nodules as compared to selected examples of real nodule cases. Taken together, our findings suggest that variations in prevalence rate can impact how readily one can learn to detect challenging examples of lung nodules in 2D chest radiographs. Our findings further suggest that the low prevalence rate of lung nodules in the clinic may have a moderately negative impact on the training of radiologists to perform this challenging diagnostic task.