Measuring the Dynamic Balance of Integration and Segregation Underlying Conscious States
Poster No:
2464
Submission Type:
Abstract Submission
Authors:
Hyunwoo Jang1, George Mashour1, Anthony Hudetz1, Zirui Huang1
Institutions:
1University of Michigan, Ann Arbor, MI
First Author:
Co-Author(s):
Introduction:
Integration and segregation are two key features of neural systems thought to mediate consciousness. Their balance in brain networks is critical for optimal function and has been a focus of research for over a decade. However, a consensus on how, precisely, to measure this balance has proven elusive. In this study we aimed to formulate a metric that captures two major surrogates of integration-segregation balance, i.e., functional connectivity strength and network topology. Through application to dynamic functional connectivity in humans, we investigate the potential of this metric to reflect shifts in conscious states during the loss and recovery of responsiveness (LOR and ROR, respectively). Furthermore, at a more granular level, we examine whether a sequential pattern of subnetwork changes in integration-segregation balance exists during the transitions to LOR and ROR. Finally, we use machine learning to identify if the topo-connectivity can reliably predict different states of consciousness.
Methods:
Our approach involved analyzing two independent fMRI datasets encompassing scans taken before, during, and after the administration of the sedative-hypnotic drug propofol. Our analysis centered on dynamic functional connectivity of 450 regions of interest, employing a sliding-window technique. The goal was to incorporate both functional connectivity strength and network topology in a single metric, to advance beyond the status quo of assessing these properties independently. Functional connectivity strength was computed through the average of the functional connectivity matrix, while topology was measured using the small-worldness metric. The product of these two was termed the "Topo-Connectivity Index". This composite index was calculated for each time window both at a whole-brain level and within distinct subnetworks. We employed various machine learning techniques and tested the model performance both within datasets and in comparison between them.
·Overview of conceptual and methodological frameworks.
Results:
Compared to conscious baseline, LOR was linked to a pronounced functional segregation in the brain. This was evidenced by a significant decrease in connectivity (p=0.0022) and a shift towards a lattice-like topology (p=0.0022). The descriminatory capability of TCI (AUC = 0.76) surpassed that of connectivity strength (AUC = 0.75) and topology (AUC = 0.69) in isolation, and other commonly used balance metrics. The whole-brain topo-connectivity exhibited a decline around the onset of LOR, continuing to decrease progressively over a span of 8 minutes. In contrast, as the transition to ROR approached, the topo-connectivity initiated a rebound 5 minutes prior to the ROR and continued to rise for 10 minutes. The analysis of brain subnetworks revealed a unique sequence of changes: the route to LOR was marked by an initial drop in topo-connectivity within unimodal networks (such as visual and somatomotor), followed by a decrease in transmodal (including fronto-parietal and default-mode) and subcortical networks. The course towards ROR was characterized by the restoration of transmodal networks, followed by the recovery of the unimodal and subcortical networks. Applying a support vector machine model, topo-connectivity emerged as a robust predictor of conscious states (accuracy ≈ 85%). Notably, our model predicted the conscious state during critical transition periods with considerable accuracy (≈ 80%), even without presenting these periods in the training data.
Conclusions:
Joint quantification of connectivity strength and network topology as a surrogate for integration-segregation balance is a novel and principled approach for assessing conscious states. Our findings also reveal a distinct temporal sequence of network changes that occur during the transition into and out of anesthetic-induced loss of consciousness. The application of the topo-connectivity and machine learning techniques demonstrates potential for objectively identifying conscious states or disorders of consciousness independently of behavior.
Brain Stimulation:
Invasive Stimulation Methods Other
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
Methods Development 2
Perception, Attention and Motor Behavior:
Consciousness and Awareness 1
Sleep and Wakefulness
Keywords:
Computational Neuroscience
Consciousness
Data analysis
FUNCTIONAL MRI
NORMAL HUMAN
Other - Anesthesia; Network Science
1|2Indicates the priority used for review