Scaling and tuning to criticality in resting-state human magnetoencephalography

This study demonstrates that applying renormalization group-inspired coarse-graining to human resting-state MEG data reveals robust scale-invariant signatures of critical dynamics, suggesting that the brain operates slightly below criticality and that these scaling laws can be used to estimate the underlying excitation-inhibition balance.

The Gist

Imagine your brain is a massive, bustling city with billions of citizens (neurons) constantly talking to each other. For decades, scientists have wondered: Is this city running in chaos, or is it running on a perfect, delicate edge of order?

This paper suggests the answer is the latter. The human brain, even when you are just sitting still with your eyes closed, operates at a "Goldilocks" point called criticality. It's not too chaotic, not too rigid, but perfectly balanced to process information efficiently.

Here is how the researchers figured this out, using simple analogies.

1. The "Zoom-Out" Game (Coarse-Graining)

Usually, when we look at brain activity, we see a chaotic mess of signals from thousands of sensors on the scalp (MEG). It's like trying to understand a traffic jam by looking at every single car individually.

The researchers used a clever trick inspired by physics called Renormalization Group (RG). Think of it like zooming out on a map:

• Step 1: You look at individual sensors.
• Step 2: You group the most connected sensors together and treat them as one "super-sensor."
• Step 3: You group those super-sensors into even bigger clusters.
• Step 4: You keep zooming out until you have just one giant cluster representing the whole brain.

If the brain were just random noise, zooming out would make the signal look like a boring, smooth Gaussian curve (a bell curve). But, the researchers found something magical: No matter how much they zoomed out, the pattern stayed the same.

2. The "Fractal" City

In math, a shape that looks the same whether you zoom in or zoom out is called a fractal (like a snowflake or a fern leaf).

The study found that brain activity is a fractal.

• The Silence: They measured how often the brain goes "silent" (no activity). As they zoomed out, the silence didn't disappear; it followed a specific, complex mathematical rule (a power law).
• The Variance: They measured how much the activity fluctuates. Instead of growing randomly, the fluctuations grew in a predictable, "self-similar" way.
• The Connections: The way different parts of the brain talk to each other remained consistent, regardless of the scale.

This is the hallmark of criticality. It means the brain is tuned to a special state where information can travel everywhere without getting stuck or exploding into chaos.

3. The "Neuronal Avalanches"

You might have heard of "avalanches" in the brain. Imagine a snowflake falling on a snowy mountain. It might trigger a tiny slide, or a massive one. In a critical system, these slides (or avalanches of brain activity) follow a specific rule: small ones are common, medium ones are less common, and huge ones are rare, but they all fit a perfect curve.

The researchers found that even when they zoomed out and grouped sensors together, these avalanches didn't change their shape. They were "invariant." This proves that the brain's critical state isn't just a local trick; it's a fundamental property of the whole system.

4. The "Tightrope Walker" (Excitation vs. Inhibition)

So, how does the brain stay on this perfect edge? The researchers used a computer model to simulate the brain. They discovered that the key is the balance between Excitation (neurons firing) and Inhibition (neurons calming down).

• Too much Excitation: The brain goes into a seizure (too chaotic).
• Too much Inhibition: The brain goes to sleep or becomes unresponsive (too rigid).
• Just Right: The brain hovers slightly below the point of chaos.

The study suggests that the specific "scaling exponents" (the numbers describing the patterns) act like a thermometer for this balance. By measuring these patterns in a non-invasive scan (MEG), we might one day be able to tell if a patient's brain is too excitable or too inhibited without needing surgery.

The Big Picture

Think of the brain not as a computer processing data line-by-line, but as a living, breathing ecosystem.

• The Analogy: Imagine a forest fire. If the trees are too far apart, the fire dies out. If they are too close, the whole forest burns instantly. But if the trees are spaced just right, the fire can spread in a controlled, self-sustaining way that covers the whole forest efficiently.
• The Finding: The human brain is that perfectly spaced forest. It operates on the edge of a "phase transition," allowing it to be flexible, adaptable, and ready to react to anything, all while maintaining a stable rhythm.

In summary: This paper shows that the resting human brain is not just "idling." It is actively maintaining a complex, fractal-like balance. By using a "zoom-out" technique, the researchers proved that this balance is a universal rule of brain function, offering a new way to measure brain health and understand how we think.

Technical Summary

1. Problem Statement

The concept of criticality—the state of a system poised at a phase transition between order and disorder—has emerged as a powerful framework for understanding brain dynamics. Evidence of criticality, such as neuronal avalanches and power-law scaling, has been observed in microscopic recordings of spiking neurons in animals (e.g., mice, rats). However, it remains unresolved whether these scaling principles govern large-scale neural activity in the human brain and how they relate to underlying neural physiology.

A major challenge is applying Renormalization Group (RG) theory to neural data. In physics, RG involves "coarse-graining" (averaging over local neighborhoods) to reveal scale-invariant properties. In the brain, the interaction structure is complex and unknown. While previous studies applied correlation-based coarse-graining to small populations of neurons, it is unclear if these signatures persist in non-invasive, large-scale human recordings like Magnetoencephalography (MEG), which represent a filtered, macroscopic view of neural activity.

2. Methodology

The authors analyzed 4 minutes of eyes-closed resting-state MEG data from 100 healthy human subjects.

• Preprocessing: MEG signals were normalized and binarized by identifying "extreme events" (fluctuations exceeding $\pm 3$ standard deviations). These events serve as proxies for neuronal spikes.
• Phenomenological Renormalization Group (PRG):
  • Instead of spatial averaging (which is difficult due to unknown connectivity), the authors employed a correlation-based coarse-graining approach.
  • Procedure: At each step $k$, the algorithm calculates the correlation matrix among the remaining variables (initially $N=273$ sensors). It pairs the most highly correlated variables, sums their activity to create a new variable, and normalizes the result.
  • This process is iterated ($k=1, 2, \dots$) until a single variable remains, effectively increasing the cluster size $K = 2^k$.
• Metrics Analyzed:
  • Activity Distribution ($Q_K$): Probability distribution of non-zero activity.
  • Probability of Silence ($P_0$): Probability of a time bin having no events.
  • Variance ($Var$): Scaling of activity variance with cluster size $K$.
  • Autocorrelation Time ($\tau_c$): How long correlations persist, scaling with $K$.
  • Covariance Eigenspectra: Scaling of eigenvalues ($\lambda$) of the covariance matrix relative to their rank.
  • Neuronal Avalanches: Size and duration distributions under coarse-graining.
• Modeling: To interpret the empirical findings, the authors used an Adaptive Ising Model. This is a network of binary neurons with excitatory interactions and a time-varying negative feedback (mimicking inhibition). The model parameters ($\beta_T$ for proximity to criticality, $c$ for feedback strength) were inferred from the MEG data.

3. Key Results

A. Robust Scaling Laws in Human MEG

The study found that resting-state MEG activity exhibits robust scale-invariant behavior under coarse-graining, consistent with a system near criticality:

1. Non-Gaussian Fixed Point: The distribution of activity ($Q_K$) converges to a fixed, non-Gaussian (exponential-like) form as cluster size $K$ increases. Independent variables would converge to a Gaussian.
2. Silence Probability: The probability of silence decays as a stretched exponential, $P_0(K) \propto \exp(-K^\beta)$, with $\beta \approx 0.82$. This deviates from the exponential decay ($\beta=1$) expected for independent variables.
3. Variance Scaling: The variance of cluster activity scales as a power law, $Var(K) \propto K^\alpha$, with $\alpha \approx 1.32$. This "intermediate" value (between 1 for independent and 2 for fully correlated variables) indicates self-similar correlations.
4. Correlation Time: The autocorrelation time scales as $\tau_c \propto K^z$ with $z \approx 0.31$, indicating that temporal correlations grow with spatial scale.
5. Eigenspectrum Scaling: The eigenvalues of the covariance matrix follow a power law $\lambda \propto (\text{rank}/K)^{-\mu}$ with $\mu \approx 0.45$. Crucially, all eigenspectra for different $K$ collapse onto a single curve when ranks are rescaled by $K$.

B. Invariance of Neuronal Avalanches

The scaling properties of neuronal avalanches (sequences of events bounded by silence) were found to be invariant under the coarse-graining procedure. The power-law exponents for avalanche size ($\tau_s$) and duration ($\alpha_t$), as well as the scaling relation between size and duration ($\langle s \rangle \propto T^\gamma$), remained stable across different levels of coarse-graining. This confirms that avalanche criticality is a fundamental property of the macroscopic brain state.

C. Relationship Between Exponents

The scaling exponents are not independent but follow specific linear relationships:

• $\beta$ decreases linearly as $\alpha$ increases.
• $\mu$ increases linearly with $\alpha$ and decreases with $\beta$.
• These relationships suggest a unified underlying mechanism governing the scaling behavior.

D. Role of Excitation-Inhibition (E/I) Balance

Using the Adaptive Ising model, the authors demonstrated that:

• The observed scaling exponents ($\alpha, \beta, z, \mu$) are reproduced only when the model operates slightly below criticality (subcritical regime).
• Feedback (Inhibition) is Critical: When the negative feedback (inhibition) was removed ($c=0$), the system lost its non-trivial scaling properties. The exponents $\beta$ and $\alpha$ approached 1 (characteristic of independent variables), and the eigenspectrum became flat.
• This implies that the E/I balance is the key physiological parameter tuning the brain to this critical-like state.

E. Validation via Surrogates

Control tests confirmed that these results are not artifacts:

• Phase Randomization: Destroying cross-correlations eliminated scaling (Gaussian distributions, $\alpha \approx 1$).
• Trace Randomization: Destroying temporal autocorrelations eliminated the scaling of $\tau_c$ but preserved other scaling, indicating $\tau_c$ depends on local dynamics while other metrics depend on collective structure.
• Random Pairing: Pairing sensors randomly (ignoring correlations) disrupted the data collapse of eigenspectra, confirming that the observed scaling relies on the specific correlation structure of the brain.

4. Significance and Contributions

1. Bridging Scales: This is the first study to demonstrate that large-scale human brain activity (MEG) obeys the same scaling laws and RG fixed-point properties previously observed only in microscopic populations of spiking neurons. This validates the universality of critical dynamics across spatial scales.
2. RG as a Tool for Human Data: The paper establishes a rigorous, correlation-based coarse-graining method to apply RG concepts to non-invasive human electrophysiology, overcoming the lack of known anatomical connectivity.
3. Physiological Insight: By linking the scaling exponents to the Excitation/Inhibition (E/I) balance via modeling, the study provides a principled route to estimate this crucial physiological parameter from non-invasive recordings. Deviations in these exponents could potentially serve as biomarkers for pathologies involving E/I imbalance (e.g., epilepsy, schizophrenia).
4. Criticality in the Human Brain: The findings provide comprehensive evidence that the human brain in the resting state operates in a critical-like regime, characterized by scale invariance, long-range correlations, and neuronal avalanches, supporting the hypothesis that criticality is a fundamental organizing principle of brain function.

In summary, the paper demonstrates that the human brain's macroscopic dynamics are governed by critical scaling laws that are robust to spatial coarse-graining, and that these laws are intimately tied to the balance of excitation and inhibition within neural networks.