Fig. 3: Neurofeedback increases β-burst power.
From: Real-time detection of neural oscillation bursts allows behaviourally relevant neurofeedback
Examples of raw LFP traces from the first (a, top) and last (b, top) of the nine neurofeedback training sessions. Time points in which beta-power exceeded the threshold defined as the 98th percentile are marked in blue (a) or red (b). Reward was delivered at time = 0. The heterogeneity of the individual bursts disappeared when averaging all bursts of one session, resulting in a sustained oscillation (a, b, bottom). c Correlation between number of rewards and session mean beta-power relative to session 1 for each rat in each session. Colours indicate the session number and each rat is denoted with a different shape. Pearson’s ρ = 0.89, p = 5.9 × 10–10. Mean power analysis of the targeted beta frequencies (20–25 Hz for rats 1 and 3 [d and j], 15–20 Hz for rat 2[g]) revealed for each rat a significant increase in power in a certain session (aha-moment) that persisted until the end of the experiment. Sessions before the aha-moment are represented in blue, and after the aha-moment in red. Two-way ANOVA (frequency and session), effect for session: rat 1: F(5,8) = 101.99, p = 3.81 × 10−24, rat 2: F(5,8) = 85.73, p = 10−22, rat 3: F(5,8) = 248.85, p = 1.29 × 10−31. ***p < 5.19 × 10−6, multiple comparisons with Bonferroni correction. Analysis of the broadband power 200 ms prior to reward delivery in a session before (blue) and after (red) neurofeedback training (e, h, k) or sham training (f, i) is plotted as mean ± 95% confidence interval (dashed). Faded lines: power 200 ms post reward delivery. Grey shading: frequencies in which the difference between power before and power after reward was significantly (p < 0.01) different after training compared to before training (t-test with Bonferroni correction). Digits (1 or 9) in the legend in e indicate the session numbers, “burst” refers to 200 ms prior to rewards and “control” to 200 ms after reward. l Group averaged beta-power change relative to the first session. Three-way ANOVA (rat, session and treatment), effect for session: trained: F(8,16) = 2.864, p = 0.0349, sham: F(8,8) = 0.385, p = 0.9. Effect for treatment: F(1,35) = 4.99, p = 0.032. **p < 0.01, *p < 0.05, multiple comparisons with Bonferroni correction. Presented elements in d, g, j, l centre line: median; box limits: upper and lower quartiles; whiskers: full distribution.
