Abstract
Introduction: There is growing interest in understanding the neural mechanisms mediating perception of natural scenes (Thorpe et al. 1996, Codispoti et al. 2006). Studies have demonstrated the presence of high-level task-related decision processes in natural scene categorization (VanRullen & Thorpe, 2001). Unlike previous studies restricted to limited categories including cars, people and animals, we can reliably detect the presence of arbitrary searched objects from neural activity (electroencephalography, EEG). Here we analyzed EEG signals using multivariate pattern classifiers (MVPC) to predict on a single trial basis the presence or absence of cued arbitrary objects during search in natural scenes. Method: Ten naive observers performed a visual search task where the target object was specified by a word (500ms duration) presented prior to a natural scene (100ms). Four hundred target present and four hundred absent images were presented. Observers used a 10-point confidence rating scale to report whether the target was present or absent. Results: The results revealed a positive deflection in the event-related potential (ERP) over parietal electrodes during the 300-700 ms post-stimulus time window that was larger for target present trials than absent trials (p<0.05). Classifier performance (area under the ROC) identifying whether the image contained the target from single trial EEG activity (ranging from 100-700 ms) was 0.69 ± 0.02 (stder). In addition, MVPC predicted reliably the trial-to-trial choices of observers (0.66 ± 0.03). Temporally windowed classifier analysis indicated that the neural activity predicting the choice of the observers was temporally distributed starting at 300 ms and sustained to 700 ms post-stimulus. Conclusion: Pattern classifiers can be used efficiently to predict the presence of arbitrary target objects in natural scenes from single trial EEG. Our analysis indicates that discriminatory neural activity is associated with late components related to observers' decision processes.