Serial dependence in the perceptual judgments of radiologists

Abstract

In radiological screening, clinicians scan myriads of radiographs with the intent of recognizing and differentiating lesions. 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 of previously seen radiographs. However, recent studies have shown that human perception is biased by previously seen stimuli; the bias in our visual system to misperceive current stimuli towards previous stimuli is called serial dependence. Here, we tested whether serial dependence impacts radiologists' recognition of simulated lesions embedded in actual radiographs. We found that serial dependence affected radiologists' recognition of simulated lesions; perception on an average trial was pulled 13% toward the 1-back stimulus. Simulated lesions were perceived as biased towards the those seen in the previous 1 or 2 radiographs. Similar results were found when testing lesion recognition in a group of untrained observers. Taken together, these results suggest that perceptual judgements of radiologists are affected by previous visual experience, and thus some of the diagnostic errors exhibited by radiologists may be caused by serial dependence from previously seen radiographs.

Keywords: Priming; Radiological screening; Sequential dependence; Sequential effects; Serial dependence; Visual search.

Fig. 1

Stimuli and design of the Experiments 1 and 2. A We created three objects with random shapes (prototypes A/B/C, shown in a bigger size) and generated 48 morph shapes in between each pair (147 shapes in total). We used these shapes as simulated lesions during radiological screening. B Observers were presented with a random shape (simulated lesion) hidden in a mammogram section, followed by a noise mask. Radiologists were then asked to adjust the shape to match the simulated lesion they previously saw, and pressed spacebar to confirm. During the inter-trial-interval, a red fixation dot appeared in the center. The size of the shape adjustment is identical to the size of the simulated lesion, but it was enlarged for illustrative purposes. After a 250 ms inter-trial interval, the next trial started

Fig. 2

Continuous Report Discrimination index (C.R.D). A For each observer, we plotted a frequency histogram of the adjustment errors and fitted a Von Mises to quantify adjustment performance. B We then converted the von Mises fit into a Cumulative Distribution Function. Continuous Report Discrimination index was calculated by taking the half difference between 25 and 75th percentile in terms of adjustment error morph units. C Each dot shows CRD index for individual observers in the two groups. Bars indicate average in Experiment 1 and 2, and error bars indicate standard error

Fig. 3

Serial dependence in the perception of simulated lesions by expert radiologists and untrained observers. A, B In units of shape morph steps, the x-axis is the shortest distance along the morph wheel between the current and one-back simulated lesion, and the y-axis is the shortest distance along the morph wheel between the selected match shape and current simulated lesion. Positive x axis values indicate that the one-back simulated lesion was clockwise on the shape morph wheel relative to the current simulated lesion, and positive y axis values indicate that the current adjusted shape was also clockwise relative to the current simulated lesion. The average of the running averages across observers (blue line) reveals a clear trend in the data, which followed a derivative-of-von-Mises shape (model fit depicted as black solid line; fit on average of running averages). Light-blue shaded error bars indicate standard error across observers. Lesion perception was attracted toward the morph seen on the previous trial. Importantly, it was tuned for similarity between previous and current morph (feature tuning). C, D The derivative-of-von Mises was converted into its source von Mises function (y-axis), and the relative morph difference was plotted in terms of CRD units (x-axis). Violet shaded error bars indicate 95% confidence interval. The curve indicates the proportion of change in response predicted by the change in the sequential stimulus. E, F Bootstrapped half amplitudes of derivative of von Mises fit for 1, 2, and 3 trials back. Half amplitude for 1-forward is shown as a comparison (grey bars). Each filled dot represents the bootstrapped half amplitude (morph units) for a single observer. Bars indicate the group bootstrap and error bars are bootstrapped 95% confidence intervals

Fig. 4

Serial dependence effect size estimation. A, B Blue lines indicate the average of the running averages across observers (same data as Fig. 2). Light-blue shaded error bars indicate standard error across observers. We fitted a linear regression on the response error as a function of the relative morph difference from − 17 to + 17 morph units (model fit depicted as green dashed line; fit on average of running averages). Dark green shaded areas indicate the morph relative difference considered in the regression analysis. C, D Bootstrapped regression slopes for 1, 2, and 3 trials back. Each filled dot represents the regression slope for a single observer. Bars indicate the group bootstrap slope and error bars are bootstrapped 95% confidence intervals

Fig. 5

Spatial tuning of serial dependence. A refers to Experiment 1, whereas B refers to Experiment 2. Each red dot refers to a different relative angular distance between current lesion and lesion in the 1-back trial, super-subject bootstrapped mean. For example, a bin distance 0° indicates that current and previous simulated tumor presented at the same location (30° of angular distance, for example). Error bars are bootstrapped 95% confidence intervals. Dashed line indicates half-amplitude zero (no bias)