Predicting creative behavior using resting-state electroencephalography

Abstract

Neuroscience research has shown that specific brain patterns can relate to creativity during multiple tasks but also at rest. Nevertheless, the electrophysiological correlates of a highly creative brain remain largely unexplored. This study aims to uncover resting-state networks related to creative behavior using high-density electroencephalography (HD-EEG) and to test whether the strength of functional connectivity within these networks could predict individual creativity in novel subjects. We acquired resting state HD-EEG data from 90 healthy participants who completed a creative behavior inventory. We then employed connectome-based predictive modeling; a machine-learning technique that predicts behavioral measures from brain connectivity features. Using a support vector regression, our results reveal functional connectivity patterns related to high and low creativity, in the gamma frequency band (30-45 Hz). In leave-one-out cross-validation, the combined model of high and low networks predicts individual creativity with very good accuracy (r = 0.36, p = 0.00045). Furthermore, the model's predictive power is established through external validation on an independent dataset (N = 41), showing a statistically significant correlation between observed and predicted creativity scores (r = 0.35, p = 0.02). These findings reveal large-scale networks that could predict creative behavior at rest, providing a crucial foundation for developing HD-EEG-network-based markers of creativity.

Fig. 1. Depictions of high and low creativity networks.

a Circle plots and (b) glass brains of high and low creativity networks. Colors within the circle plots correspond to brain lobes. b (1): the right middle temporal gyrus, (2): the lingual gyrus, (3): the left paracentral lobe, (4): the isthmus of the cingulate gyrus, (5): the rostral anterior cingulate cortex, (6): the lateral occipital cortex.

Fig. 2. The internal validation of the CPM model.

The relationship between observed and predicted creativity scores in the leave-one-out cross-validation of the CPM model, showing Pearson’s correlation coefficient (r), the p value (p), the R-squared (R²), and the Mean Absolute Error (MAE). N = 90 healthy participants.

Fig. 3. The external validation of the CPM model.

The relationship between observed and predicted creativity scores in the external validation of the CPM model, showing Pearson’s correlation coefficient (r), the p value (p), the R-squared (R²), and the Mean Absolute Error (MAE). N = 41 healthy participants, independent sample.

Fig. 4. Depictions of high creativity network of dynamic connectivity analysis.

a Circle plot and (b) glass brain of the high creativity network. Colors within the circle plots correspond to brain lobes. The network was consistent across 99% of iterations within the cross-validation loop.

Fig. 5. The internal validation of the dynamic connectivity model.

The relationship between observed and predicted creativity scores in the leave-one-out cross-validation of the CPM model, showing Pearson’s correlation coefficient (r), the p value (p), the R-squared (R²), and the Mean Absolute Error (MAE). N = 90 healthy participants.

Fig. 6. The external validation of the dynamic connectivity model.

The relationship between observed and predicted creativity scores in the external validation of the CPM model, showing Pearson’s correlation coefficient (r), the p value (p), the R-squared (R²), and the Mean Absolute Error (MAE). N = 41 healthy participants, independent sample.

Fig. 7. A full pipeline of the employed CPM method, as described in the “connectome-based predictive modeling” method section.

1 We employed CPM on the first dataset consisting of 90 healthy participants (N) to construct predictive networks that can be used to estimate individual creative behavior scores (evaluated through the Inventory of Creative Activities and Achievements or ICAA) from resting-state EEG functional connectivity. 2 the standardized weight values of edges in the functional connectivity matrix of each participant were calculated (for 68 Regions Of Interest or ROIs), where a Z-transformation was performed on each edge by calculating the difference between its weight and the mean weight divided by the standard deviation across the subjects. 3 Standardized weight values and creative behavior scores were correlated for network selection. We retained the most significant edges (p < 0.01) and grouped them into positive or negative tails. 4 Based on the selected networks, a single-subject summary index (∑) was computed for each participant (S) by summing the weight values of positive tail edges and subtracting the weight values of negative tail edges. 5 Then, a support vector regression (SVR) model was fitted to relate single-subject summary index and creativity scores. 6 The robustness of the model was assessed via an internal validation, where a Leave-one-out cross-validation (LOOCV) was applied. This strategy consists of removing one subject from the data as a novel observation, using N-1 subjects to build the predictive model, and then using the novel subject to test its prediction performance. This step is repeated N times with a different subject left out in each iteration (N times Loop). The resulting performance presents the average performance across all iterations. To assess the model’s predictive power, we evaluated the relationship between the observed and the predicted creativity scores using several metrics: Pearson’s correlation (r), parametric p-value, mean absolute error (MAE), and the coefficient of determination (R2).