Abstract
Recognizing the need for systematic large-scale visual research in the fields of psychology and neuroscience, recent work has focused on the design of large databases of images curated for human research. This raises the question: how can we collect reliable neural responses to such large numbers of images? Recent work showed that rapid serial visual presentations at 50ms per picture yield informative but compromised neural responses [Grootswagers et al. 2019]. Here we investigated how fast we can accelerate image presentation sequences without a critical impact on the neural encoding of low-level, high-level, and task-related information in MEG. We collected MEG data while participants (N=15) viewed images of 92 everyday objects presented at three rates: 500ms, 350ms, and 200ms per picture. In separate sessions, participants performed a low-level task (detection of a blank frame) and a high-level task (detection of images of female faces). We used multivariate pattern analysis with support vector machines to compare the low-level, high-level and task-related information encoded with each visual presentation sequence. We found that MEG responses to the 500ms and 350ms per picture rates contained similar information at both the single-image and categorical level. The 200ms per picture rate was largely comparable with the other rates, but had pronounced differences at the task-related level and minor differences at categorical information level. Taken together, our results show that presentation rates may be accelerated as fast as 350ms per picture with hardly any impact on the MEG neural signals.