Cortical and subcortical brain networks predict prevailing heart rate

Figures

• 

  Figure 1:. Analysis pipeline schematic.

• 

  Figure 2:. A) Prediction of heart period from physiological noise component of fMRI signal. B) Prediction of heart period after removing artifacts from the fMRI signal and accounting for the HRF delay.

• 

  Figure 3:. A) Mean out-of-sample Pearson correlation coefficient, r, of predicted vs. observed instantaneous heart period across 14 sessions (42 runs total) for each lag time shift for the original (orange) and cleaned (blue) fMRI signal. Shaded regions represent 95% confidence intervals (calculated using 1000 bootstrap iterations). Dashed vertical line represents the drive signal at lag time shift +7.5 seconds. B) Example observed and predicted instantaneous heart period across the timeseries at time shift +7.5 seconds for a representative run. C) One-sample t-test (FDR correction < 0.00005) of encoding weight maps of instantaneous heart period prediction across sessions and runs for the Human QA dataset participant at time shift +7.5 seconds. Positive weights are shown in red-yellow. Negative weights are shown in blue-green.

• 

  Figure 4:. A) Mean out-of-sample Pearson correlation coefficient, r, of instantaneous heart period for each participant for each lag time shift for the clean fMRI signal. Shaded regions represent 95% confidence intervals (calculated using 1000 bootstrap iterations). Dashed vertical lines represent the artifact signal (left) at leg time shift +1.33 seconds and the drive signal (right) at lag time shift +7.99 seconds. B) Correlation coefficients for each run for each participant at lag time shift +1.33 seconds, reflected as the leftmost dashed line in panel A. Bars represent the mean r values, error bars show the 95% confidence intervals. C) Probability map of encoding weights of instantaneous heart period prediction averaged across all participants at lag time shift +7.99 seconds. Positive weights are shown in red-yellow. Negative weights are shown in blue-green.

• 

  Figure 5:. A) Heat map of mean out-of-sample Pearson correlation coefficient values from the group analysis across participants at lag time shift +7.99 seconds, with training participants on the y axis and testing participants along the x axis. B) Breakdown of within participant model r values (heat map diagonals) and across participant model r values (heat map off diagonals). Bars represent the mean r values, error bars show the 95% confidence intervals. Gray points indicate the presence of S1, S6 or S7 in the train-test participant combination. C) Scatter plot of interparticipant euclidean distance, on the x axis and the average decoding accuracy (r values) on the y axis for between participant models.

• 

  Figure 6:. Correlation coefficient for each run from across scanner generalization models. The bar shows the mean r value, the error bars show the 95% confidence interval.