Abstract
Brain activity during mental imagery is often characterized as a reactivation of visual activity. Brain areas vary considerably in their response to qualitatively different visual stimuli, but it is currently unknown if these effects are preserved during mental imagery. To investigate this issue, we tested if the activity profile across different visually responsive brain areas remains stable when subjects imagine two qualitatively different kinds of stimuli. Specifically, we conducted a 7T fMRI experiment in which subjects viewed and imagined simple (bars and crosses) and complex (natural scene images and artwork) stimuli, and calculated signal-to-noise ratios (SNR) in individual voxels during imagery and vision. All 8 subjects of the Natural Scenes Dataset (NSD) experiment (Allen et al., 2022) took part in this additional scan session. For every vision run, there were 2 corresponding imagery runs. Significant differences in the SNR profile were observed across the two imagery runs for simple stimuli, alluding to a potential practice effect. We thus focused subsequent analyses on data from the second run only. We used an AlexNet-based encoding model to sort voxels according to their preferred network layer. We then calculated median SNR during vision and imagery and for both stimulus types as a function of network layer preference. During vision, median voxelwise SNR for simple stimuli was greater than for complex stimuli in voxels that preferred lower network layers, whereas, for voxels that preferred higher network layers, SNR for complex stimuli was greater. We observed the same trend during imagery, although the SNR mean and variance across layers was greatly reduced relative to vision. We conclude that while vision enjoys much higher SNR than imagery, the effect of stimulus type on SNR is preserved by the transformation from seen to imagined representations.