Decrease and recovery of integrated information Φ during anesthesia and sleep on human functional magnetic resonance imaging

Abstract

Integrated information theory (IIT) offers an axiomatic framework based on phenomenological properties, allowing the quantification and characterization of consciousness through a measure known as Φ. According to IIT, Φ reflects the level of consciousness and is expected to decrease with loss of consciousness, although empirical data supporting this claim remain limited. In this study, we analyzed two functional magnetic resonance imaging (fMRI) datasets acquired during anesthesia (propofol-induced) and natural sleep to determine whether Φ changes with the loss and recovery of consciousness. Our analysis was conducted using the fourth version of IIT. We constructed systems composed of five functional brain networks, computed transition probability matrices from fMRI time series data, and derived Φ values based on these matrices. As predicted by IIT, Φ decreased during anesthesia-induced loss of consciousness at both global and local levels. Similarly, Φ was locally reduced within a system centered on posterior brain regions during sleep-induced loss of consciousness. Considering functional networks as system units, we found that the integrated information (Φ) of the brain is linked to fluctuations in consciousness levels. These findings indicate a strong association between consciousness and integrated information within the large-scale functional networks.

Keywords: anesthesia; fMRI; integrated information theory; sleep.

Figure 1

(a) Analysis flow. The fMRI data were preprocessed and divided into regions based on Yeo’s atlases. Five elements were selected from the time series, and Φ was calculated for the system. Based on the binarized time series, a TPM for the states of the elements was computed. Consequently, integrated information (Φ) was calculated for each state, and the Φ values for the different levels of consciousness were compared. (b) Spatial distributions of five elements within each system of intra-network and inter-network levels. The same number or color in a distribution map indicates that the regions are included in the same element. ECN, executive control network; DAN, dorsal attention network; VAN, ventral attention network; SMN, somatomotor network; LN, limbic network; DMN, default mode network; VN, visual network.

Figure 2

Box plots illustrating the distribution of Φ across the sedation stages in the anesthesia dataset. (a) Global functional network level. (b–h) Intra-network levels in the ECN, DAN, VAN, SMN, LN, DMN, and VN, respectively. (i–l) Inter-network levels in frontal, balanced, parietal, and occipital regions, respectively. W, M, D, and R represent wake, mild sedation, deep sedation, and recovery, respectively. The boxes represent the interquartile range (IQR), with the bold horizontal lines indicating the median and the dashed lines indicating the mean. Horizontal lines above the plots denote significant differences from post hoc tests (^*P < .05, ^**P < .01, ^***P < .001, all corrected by Holm’s method for multiple comparisons).

Figure 3

Box plots illustrating the distribution of Φ across the sleep stage in the sleep dataset. (a) Global functional network level. (b–h) Intra-network levels in the ECN, DAN, VAN, SMN, LN, DMN, and VN, respectively. (i–l) Inter-network levels in frontal, balanced, parietal, and occipital regions, respectively. W, 1, 2, 3, and R denote each sleep stage (wake, N1, N2, N3, and REM, respectively). The boxes represent the IQR, with the bold horizontal lines indicating the median and the dashed lines indicating the mean. Horizontal lines above the plots denote significant differences from post hoc tests (^*P < .05, ^**P < .01, ^***P < .001, all corrected by Holm’s method for multiple comparisons).

Figure 4

Φ of baseline in the anesthesia and sleep datasets. ECN, executive control network; DAN, dorsal attention network; VAN, ventral attention network; SMN, somatomotor network; LN, limbic network; DMN, default mode network; VN, visual network.