Adaptation and the Perception of Radiological Images
StatisticsView Usage Statistics
Radiologists must classify and interpret medical images on the basis of visual inspection. We examined how an observer's visual sensitivity and perception might change as they view and thus adapt to the characteristic properties of radiological scans. Measurements were focused on the effects of adaptation to images of normal mammograms, and were tested primarily in observers who were not trained radiologists. Mammograms have steeper power spectra (slopes of ~-3) than natural images (~-2) and thus are physically blurry. Adapting to them produced shifts in the perceived spectrum of filtered noise consistent with adaptation to blur, even though this adaptation does not lead to measurable changes in the contrast sensitivity function. Strong aftereffects in the appearance of the images were also found when observers judged the perceived texture of the images. For example, tissue density in mammograms is routinely classified and ranges from "dense" to "fatty." Adaptation to dense images caused an intermediate image to appear more fatty and vice versa. Our results thus suggest that observers can selectively adapt to the properties of radiological images, and this could potentially be an important factor in the perception and learning of radiological images. In a further study we explored whether adaptation could enhance visual inspection of radiological images, specifically to aid observers in identifying abnormalities by adapting out or discounting the expected visual characteristics of the background. Observers searched for simulated lesions (Gaussian targets) added at random locations in the images. Prior adaptation to the images allowed the targets to be located more quickly, and this performance gain was selective for the tissue type and thus the visual texture defining the background. These improvements in visual search provide a novel demonstration of the advantages of spatial pattern adaptation within contexts that closely mimic routine visual tasks and settings. Finally, we explored the neural correlates of these adaptation aftereffects by measuring ERP's while observers adapted to the different textural properties of the mammogram images (dense or fatty). There was no significant difference of adapt condition in the component waveforms when tasked with categorizing the scans based upon their density classifications. In contrast, there was a significant effect of adaptation when observers were signaling target presence or absence. This significant difference was characterized by an enhancement of the neural response at early timepoints in occipital areas. Additionally, following adaptation we observed a divergence in the target present and absent waveforms at approximately 370 ms post-stimulus onset in frontal recording sites. These results suggest that target detection involves a form of the P300 component. Taken together these studies represent the first comprehensive analysis of the influence of adaption on the critically important visual judgments involved in interpreting and inspecting medical images.