Purchase this article with an account.
Joyce Sato-Reinhold, Jocelyn L. Sy, Koel Das, James C. Elliott, Miguel P. Eckstein, Barry Giesbrecht; Neural decoding during continuous task performance. Journal of Vision 2011;11(11):197. doi: 10.1167/11.11.197.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In cognitive neuroscience, multivariate pattern classification methods are typically used to discriminate between the spatial patterns of neural responses acquired using fMRI. However, these methods are also effective for capitalizing on the multivariate nature of high temporal resolution, single-trial electroencephalography (EEG) data. For instance, in perceptual decision-making tasks, EEG-based pattern classifiers can predict stimulus category and observer decisions (Philiastades and Sadja, 2006), and can outperform traditional ERP metrics in predicting individual differences in behavior (Das et al., 2010). Here we investigated whether these findings generalize to a continuous performance task. Observers viewed images of faces and cars embedded in noise (Das et al., 2010) presented in rapid serial visual presentation sequences (2 Hz) that lasted two minutes. There were four separate EEG sessions of 12,000 trials that each differed in terms of target probability (5%, 10%, 25%, 50%). A linear discriminant analysis was used to classify stimulus category (face/car) and performance (hits/misses). As in previous studies, classification accuracy discriminating stimulus type was high (mean peak classifier accuracy: 67.7%). The timecourse of this discriminatory activity peaked at approximately 300 ms post-stimulus and was strongly modulated by target probability, suggesting that the pattern classifier is likely capturing modulations in the P3 ERP component. Unlike previous studies, we found no evidence of discriminatory information about the target during the time window of the face-selective N170 ERP component. Our classifier also predicted observer performance significantly better than chance, albeit with lower success than stimulus category (mean peak classifier accuracy: 58.0%). The inclusion of EEG spectral features improved classification of performance, but not classification of stimulus type. Our results demonstrate that pattern classification algorithms can be used successfully with a continuous performance task, and that the temporal and spectral features driving classification performance are highly task dependent.
This PDF is available to Subscribers Only