Measuring the Information Content of Visually-Evoked Neuroelectric Activity
Year: 2019
Type: article
Abstract: This study addresses a fundamental question regarding the amount information contained in neuroelectric signals. More specifically, we ask the question of how many bits of information are carried by SSVEP and VEP signals regarding a particular population of natural scenes. In two experiments, observers (N=13 and N=23, respectively) viewed 150 grayscale, contrast-normalized natural scene images (subtending 18.5 deg.) while undergoing 128-channel EEG. In Experiment 1, the images were presented in a steady-state paradigm (SSVEP, 5 Hz sinusoidal contrast modulation for 6 s) and were repeated four times across the experiment. Experiment 2 was a visually-evoked potential (VEP) paradigm in which images were presented for 500 ms each and repeated six times across the experiment. For both experiments, we expressed neuroelectric responses as points in a state-space defined by the first two principal components of all responses (which capture 95%+ of the variance). If EEG signals (either SSVEP or VEP) contain reliable information about scene identity, then repeated presentations should tightly cluster in the space. We operationalized this using mutual information, which examines the relative spread of points from the same image to the spread from all points overall. Although 150 images can be expressed with 7.23 bits, we found that the SSVEP signals carried an average of 2.1 bits per image of mutual information. The VEP signals contained 1.8 bits per image when averaged across observers, and 1.0 bits when no averaging was applied, highlighting the importance of averaging to de-noise the signal. Although these conclusions are a function of the particular stimuli used in our study (normalized grayscale natural scenes) and are dependent on assumptions regarding the noise distributions, we believe our results provide an important insight into the information present in this non-invasive signal, and offer a framework for characterizing the information content of other imaging modalities.
Source: Journal of Vision
DOI: 10.1167/19.10.48c
Institutions: Bates College, Cornell University, Colgate University
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $1,850
OpenAlex ID: W2974089426