Fig. 5: Cortical state predicts behavioral outcome.

a Relationship between ongoing PR and stimulus evoked response for one example electrode during the detection task. Stimulus evoked response is positively correlated with pre-stimulus ongoing PR (Pearson’s correlation = 0.57, p < 1.0e-9). b The Pearson’s correlation probability density function for PR and evoked response across all subjects and electrodes in both tasks (0.3 ± 0.01, mean ± SEM; p = 6.51e-33, one sided Wilcoxon signed-rank test for positive median). c Population distribution of the difference in evoked responses to the target and test stimuli (test – target) for low (blue) and high (red) PR trials. Median PR value during a session was used to divide trials into low and high PR groups. The difference was significantly larger for the low PR trials than for the high PR trials (0.21 ± 0.02 dB vs. −0.09 ± 0.02 dB, mean ± SEM; p = 0.02, tow-sided Wilcoxon signed-rank test) (d) Fisher-Linear Discriminant Analysis (F-LDA) for the same pair of example electrodes during detection and discrimination tasks. Each dot represents the ongoing PR during a correct (blue) or incorrect (red) trial. The green dotted lines represent the decision boundaries of the trained F-LDAs for the two tasks. Histograms are generated by projecting each PR value onto the F-LDA axis so that the separability between correct and incorrect histograms is maximized. The curves around histograms represent one-dimensional Gaussian fits for the respective histograms. e Population average of the F-LDA performance during the detection (blue) and discrimination (red) tasks. Performance is significantly above the chance level (gray) for both detection (67.38 ± 6.39, mean ± SEM; p = 0.003, bootstrapping with 10,000 iterations) and discrimination (61.42 ± 4.7, mean ± SEM; p = 0.007, bootstrapping with 10,000 iterations) tasks. The F-LDA was trained to predict whether a subject was going to be correct or incorrect in a trial based on the pre-stimulus ongoing PR values. **p < 0.01.