Measuring robust functional connectivity from resting-state MEG using amplitude and entropy correlation across frequency bands and temporal scales

Figures

• 

  Fig. 1. Valid connections (mean rank >0.8) found using AEC correlation for four frequency bands (above) and MRVE correlation for a range of time scales (below). Each point represents an AAL region and each line represents a connection. The midpoint of the frequency band (for AEC) or scale frequency (for MRVE correlation) is indicated in the top left corner of each plot, in Hz. The key at the top indicates the colour of the connections that originate in each brain region. No valid connections were found using AEC in the 3–8, 40–60, 60–80, 80–100 or 140-160Hz frequency bands.

• 

  Fig. 2. A) The number of valid connections found for each scale frequency using MRVE correlation and each frequency band using AEC. B) The highest mean rank of each connection across all frequency bands vs. all scale frequencies, where the colour indicates whether the connection is detected by both MRVE correlation and AEC, detected by neither, or detected by only one of the methods. Circle plots show the connections that are only detected as valid by either MRVE correlation or AEC. C) Valid connections plotted on a template brain for MRVE scale frequencies fS=75Hz and 10Hz and AEC frequency bands 1-4Hz, 8-13Hz and 13-30Hz.

• 

  Fig. 3. Analysis of the robustness and inter-participant consistency of the connectivity measures given by AEC and MRVE correlation. A) Robustness to reducing the number of subjects included in the analysis as measured by bootstrapping. Plot shows the mean number of valid connections detected over 1000 sub-samples of size N, randomly sampled with replacement. Error bands show the standard deviation. B) Consistency across subjects found by correlating the vectorised z score connectivity matrices of individual subjects with the average connectivity pattern across all subjects. Colour plots show the resultant pattern-correlation coefficients for each subject, sorted by correlation strength, for each MRVE scale frequency and AEC frequency band. AEC bands are represented by the frequency at the midpoint between the limits of the frequency range. C) These pattern-correlation coefficients over subjects were transformed to z scores and averaged for each scale frequency and frequency band. Error bands show the standard deviation over 1000 sub-samples of N=90. D) Sorted pattern-correlation coefficients, calculated as in B) for MRVE scale frequency fS = 75Hz and beta band AEC. E) Cross-correlation…

• 

  Fig. 4. (left) The maximum adjusted R2 obtained across all possible multiple linear regression models for each scale frequency. (right) Colour plot showing regression coefficients, where each column represents the coefficients obtained using MRVE correlation at the given scale frequency as the response variable. AEC bands are represented by the frequency at the midpoint between the limits of the frequency range. Black indicates that the corresponding AEC frequency was not included as a predictor variable in the optimal regression model (maximising adjusted R2).

• 

  Fig. 5. The temporal correlation between MRVE timecourses and oscillatory amplitude envelopes for scale frequencies fS=1-150Hz and frequency bands 1-4Hz (δ), 3-8Hz (θ), 8-13Hz (α), 13-30Hz (β), 40-60Hz (γ40−60), 60-80Hz (γ60−80) and 140-160Hz (γ140−160). The temporal correlation coefficient was found at each voxel for each participant and transformed to a z-score by applying the Fisher transformation. The 95% confidence interval was found for the z-scores calculated across all participants for each voxel. Average Pearson correlation values were found at each voxel where z=0 lay outside of this confidence interval and displayed on a template brain as indicated by the colour bar. See supplementary material for whole brain correlation images for all scales and frequency bands.

• 

  Fig. 6. The correlation between average oscillatory amplitude/entropy and the overall connectivity strength at each voxel as measured by AEC and MRVE correlation. AEC bands are represented by the frequency at the midpoint between the limits of the frequency range. Warm colours indicate positive correlation whereas cooler colours show negative correlation and grey indicates a non-significant relationship (α=0.05, FDR corrected for multiple comparisons).

• 

  Fig. 7. A) Connections where the EOG standard deviation could explain a significant amount of variance in the connectivity strength across participants (p<.05). Red indicates a positive relationship and blue indicates a negative relationship. Asterisks mark connections that show a significant relationship after Bonferroni correction. The AEC frequency band or MRVE scale frequency is given in the top left corner of each plot. B) The number of connections showing a significant relationship. The number of connections exhibiting a positive relationship is given above the x-axis. The number showing a negative relationship are given below the x-axis. The plot above shows the number before correction for multiple comparisons. The plot below shows the number after Bonferroni correction. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)