A practical measure of integrated information reveals alpha-band activity and the posterior cortex as neural correlates of arousal.

A new measure called Φcopula, which uses a Gaussian copula approach to estimate integrated information, outperforms common estimators by maintaining the lowest bias and mean squared error even in non-Gaussian high-dimensional systems. Applied to electroencephalographic data across awake, propofol-induced unresponsive, and NREM sleep states, alpha-band Φcopula significantly decreased during both anesthesia and sleep. Φcopula-based classifiers distinguished arousal states more accurately than functional connectivity and network efficiency measures. The dorsal attention network and default mode network contributed most to Φcopula, with the cingulate and posterior cortices showing the greatest contributions. The posterior cortex, especially the posterior cingulate cortex, appears critical for arousal-related information integration and consciousness.

Study at a glance

Abstract

The search for neurophysiological markers of consciousness and their neural substrates remains a focal point in neuroscience research. The integrated information theory (IIT) provides a promising quantitative framework for consciousness assessment, but computational limitations of existing Φ estimation methods hinder an in-depth understanding of large-scale cortical integration. Here, we proposed a new measure, Φcopula, by incorporating the Gaussian copula approach for estimating integrated information. Simulation analysis demonstrated that Φcopula significantly outperformed common estimators, maintaining the lowest bias and mean squared error (MSE) even in non-Gaussian high-dimensional systems. We applied Φcopula to electroencephalographic data across different arousal states: awake, propofol-induced unresponsiveness, and non-rapid eye movement (NREM) sleep. Results revealed that alpha-band Φcopula significantly decreased during both propofol anesthesia (p < 0.001) and sleep (p < 0.014) states. Moreover, classification analysis demonstrated that Φcopula-based classifiers achieved superior accuracy in distinguishing arousal states compared to functional connectivity and network efficiency measures (p < 0.030 for anesthesia; p < 0.043 for sleep). Among the functional networks, the dorsal attention network (DAN) and default mode network (DMN) contributed most to Φcopula. Among the anatomical brain regions, the cingulate and posterior cortices showed the greatest contributions. Our findings suggest that Φcopula is a practical and effective metric for quantifying integrated information, with substantial potential for monitoring arousal levels in clinical and experimental settings. The posterior cortex, especially the posterior cingulate cortex (PCC), shows the greatest contribution to arousal-related information integration, revealing its critical role in consciousness.