Abstract
Rapid extraction of global structural regularities provides us with basic information of our visual world, the so-called gist. Scene research is divided as to whether it is low or high frequencies that are essential for driving gist signals. However, recent work has suggested that high rather than low spatial frequencies are vital for radiologists to extract the gist of abnormality from mammograms (Evans et al., 2016). Here, we investigated the effects of high-pass filtering of mammograms on detecting the gist of the abnormal. Twenty-eight radiologists viewed mammograms for 500 ms, rating their gist impression of abnormality in unaltered and filtered images. Performance on 90 no cancer, 30 obvious, 30 subtle cancer, and 30 three years prior to sign of abnormality unaltered mammograms was compared to contrast-normalized, 2nd order Butterworth high-pass filtered versions of the same (0.5, 1, 1.5 & 2 cycles per degree (cpd)). Filtering out frequencies lower than 0.5 and 1 cpd significantly affected the radiologists’ performance (ηp2=.45) with an increased performance when frequencies below 0.5 cpd were filtered out compared to both unfiltered and 1 cpd. The beneficial effect of filtering was especially pronounced for mammograms that did not have a localized signal present (ηp2=.13), with improvement from d’=0.02 for unfiltered to d’=0.71 for 0.5 filtered, staying unchanged for 1 cpd. Filtering out of frequencies below 1.5 and 2 cpd showed no discernible pattern of promoting or impeding gist extraction in general or for different conspicuities. Overall, categorization performance was higher for images with obvious cancer lesions as opposed to subtle or mammograms acquired prior to any visible actionable signs of cancer (ηp2=.59, ηp2=.78). In conclusion, our results support the idea that high frequencies can carry important information for gist extraction of abnormality, suggesting that this signal might be especially present in frequencies between 0.5 and 1 cpd.