Assessment of Coupled Phase Oscillators-Based Modeling in Swine Brain Connectome

⚕️

This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

The Big Picture: Building a "Digital Twin" of a Pig's Brain

Imagine you have a giant, complex city (the brain) with millions of roads (structural connections) and traffic patterns (functional activity). The scientists wanted to build a computer simulation—a "Digital Twin"—that could predict how traffic flows just by looking at a map of the roads.

In this study, they used pigs because their brains are very similar to human brains. They wanted to see if they could use a mathematical model to understand how the brain works normally, and then see how that model breaks down when the brain gets injured (Traumatic Brain Injury, or TBI).

The Main Characters: The "Dancers" and the "Wires"

To understand the brain, the researchers used a concept called the Kuramoto Model. Here is the best way to visualize it:

The Oscillators (The Dancers): Imagine 60 different groups of dancers standing on a stage. Each group represents a specific area of the brain. Every group has its own natural rhythm (some dance fast, some slow).
The Structural Connectivity (The Wires): These dancers are connected by invisible wires. If two groups are connected by a thick wire, they can easily see and hear each other. If the wire is thin or missing, they can't communicate well.
The Functional Connectivity (The Dance): This is what happens when they actually start dancing together. Do they all move in perfect unison? Do they form small circles? Or do they dance chaotically?

The Challenge: In the real brain, just because two areas are connected by a wire doesn't mean they will dance together perfectly. Sometimes they dance together even without a direct wire. The scientists wanted to build a computer program that could take the "Map of Wires" and accurately predict the "Dance."

The Experiment: Tuning the Radio

The scientists tried to tune their computer model to match the real pig brains. They had to adjust two main "knobs":

Natural Frequency: How fast each dancer wants to move on their own.
Coupling Strength: How hard the dancers try to pull each other into sync via the wires.

The Discovery:
They found that if they just guessed these settings, the simulation was a mess. But, when they carefully "tuned" the knobs using data from healthy pigs, the computer simulation started to look exactly like the real brain scans.

The Result: The simulated dance matched the real dance 61% of the time (a very strong match in this field). This proved that the "wires" (anatomy) are the main reason the brain dances the way it does.

The Test: What Happens When the Brain Gets Hurt?

Next, they wanted to see if their model could handle a disaster. They simulated a Traumatic Brain Injury (TBI) in pigs. They hit some pigs lightly (mild TBI) and some hard (severe TBI), then scanned them at different times: 1 day, 63 days, and 119 days after the injury.

The Findings:

Immediate Aftermath (Day 1): Surprisingly, the model still worked well immediately after the injury. The "dance" didn't change much right away. It's like if you broke a few streetlights in a city; the traffic patterns might still look normal for a little while because the drivers are compensating.
Long-Term (Months Later): Over time (63 and 119 days), the connection between the "wires" and the "dance" started to weaken. The model found it harder to predict the brain's activity.
- Analogy: Imagine a city where the roads are still there, but the drivers have changed their habits. They are taking different routes or driving differently because of the damage. The map (structure) is the same, but the traffic flow (function) has evolved.
Severity Didn't Matter: Interestingly, it didn't matter if the pig was hit lightly or hard. The model struggled in the same way for both groups over time. This suggests that the brain's long-term recovery process is a complex, shared experience, regardless of how bad the initial hit was.

What Did They Learn? (The Takeaway)

Structure is King (but not the only King): The physical wiring of the brain is the biggest factor in how it functions. If you know the roads, you can mostly predict the traffic.
The Model is a Good Tool: This computer model is a great "Digital Twin." It can mimic a healthy brain very well and can detect when the brain is starting to change after an injury.
Recovery is a Journey: The brain doesn't just "snap back" to normal after a TBI. It slowly reorganizes itself over months. The model helps us see that this reorganization changes how the brain's parts talk to each other.

Why Does This Matter?

Think of this model as a crystal ball for brain injuries.

Right now, if a human gets a TBI, doctors look at scans and guess how they will recover.
In the future, this kind of model could help doctors simulate: "If we give this patient this specific therapy, how will their brain's 'dance' change?"

It turns the brain from a mysterious black box into a system we can understand, tune, and hopefully, heal.

1. Problem Statement

The core challenge addressed in this study is the difficulty in linking Structural Connectivity (SC) (the physical white matter wiring of the brain) to Functional Connectivity (FC) (the statistical synchronization of neural activity). While SC is relatively stable, FC is dynamic and task-dependent. Empirical data often shows a weak direct correlation between the two; regions with strong structural links may not always synchronize, and vice versa.

The authors aim to bridge this gap using mechanistic models to simulate how anatomical constraints give rise to emergent functional dynamics. Specifically, they seek to:

Validate the Kuramoto phase-oscillator model in a swine (pig) connectome.
Determine optimal model parameters (natural frequencies and global coupling strength) to reproduce empirical resting-state FC.
Assess the model's sensitivity to Traumatic Brain Injury (TBI) and its ability to track longitudinal changes in structure-function coupling post-injury.

2. Methodology

Data Acquisition and Preprocessing

Subjects: 44 Landrace crossbreed pigs (8 weeks old) divided into three groups: Sham ( $n=12$ ), Mild TBI ( $n=16$ ), and Severe TBI ( $n=16$ ).
TBI Induction: Controlled cortical impact (CCI) at 4 m/s with penetration depths of 3 mm (mTBI) or 9 mm (sTBI). Sham groups underwent craniectomy only.
Imaging: Longitudinal MRI data collected at four time points: -1 day (baseline), +1 day, +63 days, and +119 days post-injury.
- dMRI: Used for tractography to construct SC matrices.
- rs-fMRI: Used to derive empirical Functional Connectivity (eFC) matrices.
Network Construction: A predefined atlas was parcellated into 60 Regions of Interest (ROIs). SC and FC matrices were normalized and symmetrized.

Modeling Framework

The study employed a structurally constrained Kuramoto model coupled with a hemodynamic model:

Kuramoto Dynamics: The phase evolution of $N$ oscillators (ROIs) is governed by:
$\frac{d\theta_i}{dt} = \omega_i + K \sum_{j=1}^{N} C_{ij} \sin(\theta_j(t-\tau_{ij}) - \theta_i(t))$
Where $\omega_i$ is the natural frequency, $K$ is the global coupling constant, $C_{ij}$ is the structural connection strength, and $\tau_{ij}$ represents conduction delays based on fiber length.
Hemodynamic Transformation: The simulated oscillator phases were converted into BOLD signals using the Balloon-Windkessel model to mimic fMRI data.
Parameter Optimization (Joint Tuning):
- Instead of using random or fixed parameters, the authors developed a joint optimization algorithm.
- They iteratively adjusted the global coupling constant ( $K$ ) and the vector of natural frequencies ( $\omega$ ) to maximize the similarity (Pearson correlation) and minimize the Mean Squared Error (MSE) between the Simulated FC (sFC) and Empirical FC (eFC).
- Two tuning strategies were tested: Individual (per pig) and Averaged (across the control group).

Validation and Analysis

Null Model Testing: Compared sFC generated from empirical SC against sFC generated from randomized SC (preserving density/degree) to ensure results were not due to chance.
Graph Theory: Evaluated topological features (Global Efficiency, Characteristic Path Length, Modularity, Clustering Coefficient, Small-Worldness) across network densities of 30–50%.
Longitudinal Assessment: Applied the tuned model (using baseline control parameters) to predict FC at post-injury time points to evaluate structure-function coupling degradation.

3. Key Contributions

Data-Driven Parameter Calibration: Demonstrated that natural frequencies cannot be assumed random; they must be empirically tuned. The study found that random frequencies failed to reproduce FC, whereas data-informed tuning significantly improved model fidelity.
Hybrid Optimization Strategy: Proposed a "Combined" (Combn) scheme using a global coupling constant ( $K=29$ ) derived from the averaged control group, paired with pig-specific natural frequencies. This approach successfully balanced the reproduction of common network patterns while preserving inter-subject heterogeneity.
Swine Connectome Modeling: Successfully established a Kuramoto-based framework for the swine brain, validating it as a viable translational model for human TBI studies.
Longitudinal TBI Tracking: Provided a quantitative framework to track the evolution of structure-function decoupling over time following TBI.

4. Key Results

Model Performance

Correlation: The optimized model achieved a significant edge-wise correspondence between averaged simulated and empirical FC ( $r = 0.61$ , $p < 0.001$ ).
Optimal Parameters:
- Individual Tuning: Optimal $K = 22$ .
- Averaged Tuning: Optimal $K = 29$ .
- Combined Scheme: Used $K=29$ with individual frequencies, yielding the highest prediction scores.
Null Model: Simulations using randomized structural networks showed significantly lower similarity to empirical FC than those using empirical SC, confirming the model captures genuine structure-function relationships.

Graph-Theoretical Analysis

High Agreement: The model accurately reproduced Global Efficiency (GE), Characteristic Path Length (CPL), and Clustering Coefficient (CC). This suggests these integrative metrics are strongly constrained by static anatomical wiring.
Deviation: The model showed poorer agreement for Modularity (MOD) and Small-Worldness (SW). These higher-order features likely depend on dynamic, state-dependent processes not fully captured by static SC alone.

Longitudinal TBI Findings

Acute Phase (+1 day): Structure-function coupling remained relatively stable across all injury severities, suggesting immediate functional compensation or that acute structural damage does not immediately disrupt large-scale coordination.
Chronic Phase (+63 and +119 days): A modest but progressive reduction in structure-function coupling was observed over time for all groups.
- This decline is attributed to evolving changes in white matter integrity and shifts in intrinsic oscillatory properties (natural frequencies) that the static model (tuned on baseline data) could not fully capture.
Injury Severity: No significant differences were found in model performance between Sham, mTBI, and sTBI groups at any time point, indicating the model's robustness across injury severities.

5. Significance and Conclusion

Mechanistic Insight: The study confirms that while structural connectivity is a primary driver of functional organization, accurate simulation requires precise calibration of regional dynamics (natural frequencies).
Translational Potential: The framework provides a quantitative tool to probe how TBI reshapes brain dynamics. The observed decoupling of structure and function over time serves as a potential biomarker for maladaptation or incomplete recovery.
Limitations & Future Work: The study notes limitations in sample size and the 4-month follow-up duration. Future work aims to extend longitudinal optimization to track injury-specific dynamical changes in real-time, potentially guiding therapeutic interventions.

In summary, this paper establishes a robust, data-driven Kuramoto modeling framework for the swine brain, demonstrating its utility in reproducing functional networks and sensitively tracking the longitudinal degradation of structure-function coupling following traumatic brain injury.