Decreased Cross-Domain Mutual Information in Schizophrenia From Dynamic Connectivity States

Mustafa S. Salman, Victor M. Vergara, Eswar Damaraju, Vince D. CalhounFrontiers in Neuroscience 2019 · DOI

Abstract

The study of dynamic functional network connectivity (dFNC) is essential for understanding both healthy and diseased brain states. This research introduces Dynamic Functional Domain Connectivity (DFDC), a novel information-theoretic framework designed to analyze the temporal sequence of brain states and the information shared among functional domains. Applying this framework to resting-state fMRI data from 163 healthy controls (HCs) and 151 schizophrenia patients (SZs), the study measured domain state probabilities, entropy, and cross-domain mutual information (CDMI). Results indicate that SZs exhibit significantly higher entropy in specific domain pairs and lower CDMI across multiple networks. These findings suggest that schizophrenia is characterized by higher uncertainty, randomness, and fragmented communication within the brain's functional architecture.

Introduction: Schizophrenia as a Disconnection Syndrome

Schizophrenia has long been conceptualized as a "disconnection syndrome," where the primary deficit lies in the breakdown of communication between brain regions. While traditional fMRI studies often rely on static snapshots of brain activity, the brain is inherently dynamic. This paper addresses the gap in understanding how groups of related brain regions—functional domains—interact over time. By treating the brain as a digital communication network, the authors quantify the efficiency, predictability, and coordination of information flow to determine if the information-theoretic signature of schizophrenia is fundamentally more disorganized than that of a healthy brain.

Methodology: The DFDC Framework

The researchers utilized the Dynamic Functional Domain Connectivity (DFDC) framework to quantify neural information flow. The workflow, illustrated in Figure 1, involves several sophisticated stages:

  1. Group Independent Component Analysis (ICA): Decomposing fMRI data into 47 distinct functional networks.
  2. Domain Grouping: Organizing networks into domains such as Subcortical (SC), Sensorimotor (SM), Visual (VIS), Attention (ATTN), Frontoparietal (FRN), and Default Mode (DMN).
  3. Sliding Window Analysis: Using moving time windows to observe fluctuations in connectivity.
  4. K-means Clustering: Identifying recurring "connectivity states" for each domain pair.
  5. Information Theory Metrics: Calculating Entropy (a measure of uncertainty or randomness in state transitions) and Cross-Domain Mutual Information (CDMI) (a measure of coordinated activity across different domain interactions).
The Information Theoretic Framework of Brain Connectivity workflow.

Key Results: Increased Randomness and Reduced Coordination

The study utilized a large dataset from the FBIRN phase-III study. The results highlight a clear distinction in how information is processed in the schizophrenic brain compared to healthy controls.

The "Noisy" Brain

Patients with schizophrenia showed significantly higher transformed entropy in several domain pairs, most notably Subcortical-Subcortical (SC-SC), Default Mode-Visual (DMN-VIS), and Frontoparietal-Visual (FRN-VIS). High entropy indicates that brain states transition in an unpredictable, "noisy" fashion, suggesting that neural activity in SZs is less settled into stable patterns. Even at the whole-brain level, dFNC entropy was significantly higher in the schizophrenia group ($p = 0.0025$).

Fragmented Communication

A core finding was the decrease in CDMI among patients with schizophrenia. In healthy brains, different domain pairs (e.g., Visual-Subcortical and Sensorimotor-Subcortical) exhibit synchronized activity, sharing significant information. In contrast, these pairs function more independently in schizophrenia, suggesting a breakdown in global brain coordination. Figure 2 illustrates the relationship between these state probabilities and the clinical diagnosis.

Mean state probabilities and their relationship to diagnosis (HC vs. SZ).

The authors hypothesize that subcortical regions, such as the thalamus, may be "overworked" in schizophrenia, relaying excessive or noisy signals to the cortex. This manifests as higher entropy but lower meaningful coordination. Interestingly, the decrease in information sharing between certain visual and cognitive networks correlated with the severity of "positive" symptoms, such as hallucinations and delusions.

Conclusion

This study provides evidence of higher uncertainty and randomness in schizophrenic brain function. By applying an information-theoretic lens, the researchers demonstrated that functional units in the SZ brain operate in a more isolated and unpredictable manner. While future work is needed to refine cluster selection and test these metrics across other psychiatric disorders, the DFDC framework offers a promising new approach to identifying biomarkers for mental illness.

Summary updated

Abstract

The study of dynamic functional network connectivity (dFNC) has been important to understand the healthy and diseased brain. Recent developments model groups of functionally related brain structures (defined as functional domains) as entities that can send and receive information. A domain analysis starts by detecting a finite set of connectivity patterns known as domain states within each functional domain. Dynamic functional domain connectivity (DFDC) is a novel information theoretic framework for studying the temporal sequence of the domain states and the amount of information shared among domains. In this setting, the information flow among functional domains can be compared to the flow of bits among entities in a digital network. Schizophrenia is a chronic psychiatric disorder which is associated with how the brain processes information. Here, we employed the DFDC framework to analyze a dataset containing resting-state fMRI scans from 163 healthy controls (HCs) and 151 schizophrenia patients (SZs). As in other information theory methods, this study measured domain state probabilities, entropy within each DFDC and the cross-domain mutual information (CDMI) between pairs of DFDC. Results indicate that SZs show significantly higher (transformed) entropy than HCs in subcortical (SC)-SC; default mode network (DMN)-visual (VIS) and frontoparietal (FRN)-VIS DFDCs. SZs also show lower (transformed) CDMI between SC-VIS vs. SC-sensorimotor (SM), attention (ATTN)-VIS vs. ATTN-SM and ATTN-SM vs. ATTN-ATTN DFDC pairs after correcting for multiple comparisons. These results imply that different DFDC pairs function in a more independent manner in SZs compared to HCs. Our findings present evidence of higher uncertainty and randomness in SZ brain function.
BibTeX
@inproceedings{salman2019decreased,
  author = {Mustafa S. Salman and Victor M. Vergara and Eswar Damaraju and Vince D. Calhoun},
  title = {{Decreased Cross-Domain Mutual Information in Schizophrenia From Dynamic Connectivity States}},
  booktitle = {Frontiers in Neuroscience},
  year = {2019},
  doi = {10.3389/fnins.2019.00873},
}
Figure