Abstract
Radiologists can identify the gist of a medical image (abnormal vs. normal) better than chance in static 2D images, after presentations of half a second or less (Evans et al., 2013; 2016). The gist of 2D real-world scenes is carried by low spatial frequency channels, which convey the structural layout of scenes (Schyns & Oliva, 1994). In contrast the gist signal in 2D mammography is carried by high spatial frequency channels (Evans et al. 2016). Standard practice in radiology is moving to 3D modalities, where each case consists of a series of images that are assembled into a virtual stack. Radiologists can extract gist from movies of these stacks (Trevino et al., 2019; Wu & Wolfe, 2019). We do not know which channels carry the gist from 3D stacks. We tested 51 radiologists with prostate mpMRI experience on 56 cases, each comprising a stack of 26 T2-weighted prostate mpMRI images. Lesions (Gleason scores 6-9) were present in 50% of cases. A trial consisted of a movie of a single case presented at 48 ms/slice. After each case, participants localized the cancerous lesion on a prostate sector map, then indicated whether a cancerous lesion was presented, and gave a confidence rating. Radiologists were divided equally into three groups who viewed either unfiltered images, low-pass (< 2 cycles/°) filtered images, or high-pass (> 6 cycles/°) filtered images. Unfiltered detection and localization performance were higher than chance (d’ = 0.28; localization = 31%). Radiologists performed at chance when detecting lesions in high-pass filtered images (d’ = 0.16), and were significantly lower than chance for low-pass filtered images (d’ = -0.23). Our data indicate that gist perception from 3D prostate MRI relies on spatial frequency channels between 2 and 6 cycles/°. These findings emphasize that scene gist is highly dependent on task and context.