Purchase this article with an account.
Malerie G McDowell, William R Winter, Edwin F Donnelly, Frank Tong; Training with simulated lung nodules in X-rays can improve the localization performance of radiology residents. Journal of Vision 2019;19(10):27c. doi: 10.1167/19.10.27c.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
It is estimated that a certified radiologist will have read between 10,000 and 20,000 chest radiographs upon completing a four-year training residency and clinical fellowship. Despite extensive exposure to chest radiographs, residents rarely have the opportunity to read cases involving solitary lung nodules because of their low incidence rate of ~0.2%. Inadequate experience with lung nodule cases may contribute to their high rate of missed detection, which ranges from 20–30% in retrospective clinical studies. We sought to investigate whether a regimen of visual-cognitive training at a challenging lung nodule localization task might improve localization performance for these nodules. Due to the potential training relevance of this study, radiology residents were recruited as observers for this experiment. Because of the low incidence rate, lung nodules in our training stimuli were simulated using lab-developed software, which allowed control over their location, contrast, size and spatial profile. Residents (n=6) were asked to localize the occurrence of a nodule in individual chest X-rays over 4 training sessions (150 trials per session). We also obtained pre-training and post-training measures of performance, by presenting 20 cases of simulated nodules and also 20 cases of real nodules to evaluate whether benefits of training would generalize. During training sessions, audio and visual feedback was given to indicate the correct location of the nodule in each trial. This training resulted in increased localization performance wherein accuracy in detecting simulated nodules was significantly increased (52.5% pre-test, 86.7% post-test, p = 0.004), and a non-significant trend was observed for real nodules (62.5% pre-test, 71.7% post-test, t(5) = 1.53, p = 0.19). The data gathered so far suggest that, while our simulation and training methods may be further improved, this paradigm of nodule localization training may have the potential to improve clinical performance in nodule detection.
This PDF is available to Subscribers Only