Measuring Functional Connectivity Through Phase Synchronization

38 concepts34 papers1 proposal

To understand how the brain processes information, neuroscientists traditionally looked at the "volume" of neural activity—the amplitude or spectral power of localized brain waves. If a brain region showed high-amplitude oscillations in a specific frequency band, it was assumed to be actively engaged in a task.

However, the brain is a highly distributed network. Complex cognitive functions, from executing a motor command to maintaining alertness, require precise communication between geographically separated cortical regions. Amplitude alone cannot tell us how these regions interact. To map this internal communication, researchers look instead at phase synchronization: the precise timing relationships between neural oscillations.

Beyond Amplitude: The Case for Timing

Why prioritize phase over amplitude? Imagine a global logistics network. Measuring amplitude is like counting how many trucks are parked at individual warehouses. It tells you local capacity, but it doesn't tell you if the warehouses are successfully coordinating deliveries. Phase synchronization measures the traffic schedule. When two brain regions are phase-synchronized, their electrical oscillations rise and fall in tandem, opening a predictable window for information transfer.

Early evidence of this mechanism's importance came from studies on cognitive states. Scott Makeig et al. demonstrated that drops in human alertness during detection tasks are tightly coupled with shifts in spectral coherence and the emergence of phase lags between distinct scalp sites, rather than simple changes in isolated signal power. Similarly, when decoding intentions for Brain-Computer Interfaces (BCIs), Le Song et al. found that the relative timing (phase) of brain waves during the first two seconds of a task provided more discriminative power for motor imagery than traditional amplitude-based metrics.

By shifting focus from local power to inter-regional synchronization, researchers can construct a functional connectivity network that maps how the brain coordinates its distributed resources.

Signal A:  /\  /\  /\  /\  (Phase and amplitude steady)
Signal B:  /\  /\  /\  /\  (In phase with A -> High Synchronization)
Signal C:  \/  \/  \/  \/  (Shifted but constant phase relation -> High Synchronization)
Signal D:  /\    /\  /     (Drifting phase relation -> Low Synchronization)

Quantifying Alignment: The Phase-Locking Value

To mathematically capture this timing coordination without being misled by fluctuations in signal strength, researchers widely use the Phase-Locking Value (PLV).

The core intuition behind PLV is simple: strip away the amplitude of two signals entirely, isolate their instantaneous phases over time, and measure how constant their phase difference remains.

  1. Phase Extraction: For two raw time-series signals (such as EEG channels), a mathematical transform (typically the Hilbert transform) extracts the instantaneous phase, $\theta(t)$, at every time step for a specific frequency band.
  2. Phase Difference: The algorithm calculates the difference between these two phases: $\Delta\theta(t) = \theta_1(t) - \theta_2(t)$.
  3. Vector Averaging: Each phase difference is represented as a vector of length 1 on a complex unit circle ($e^{i\Delta\theta(t)}$). The vectors across all time points are then averaged.

$$\text{PLV} = \frac{1}{N} \left| \sum_{t=1}^{N} e^{i\Delta\theta(t)} \right|$$

If the phase difference varies randomly from moment to moment, the vectors will point in all directions around the circle, canceling each other out and yielding a PLV near $0$. If the phase difference remains perfectly locked—meaning the two regions maintain a rigid timing relationship—the vectors will all point in the same direction, summing to a PLV of $1$.

Because PLV normalizes the signal amplitude to a unit vector, it is exceptionally robust to transient bursts of energy. For instance, Thomas G. Simpson et al. utilized PLV to track brief, transient "beta bursts" in the motor cortex, demonstrating how these short-lived oscillatory events successfully synchronize cortical networks and peripheral muscles to maintain stable postures.

This mathematical properties make PLV a versatile proxy for functional connectivity across diverse domains, from tracking network changes during mindfulness meditation (Asieh Ahani et al.) to mapping emotional processing in response to visual stimuli (Zhongyang He et al.).

The Volume Conduction Problem and the Phase Lag Index

While PLV is powerful, it suffers from a major physical vulnerability when applied to scalp recordings like EEG: volume conduction.

The brain tissue, cerebrospinal fluid, and skull act as electrical conductors. A single, strong electrical source deep within the cortex can radiate outward and register simultaneously on multiple nearby scalp electrodes.

       [ Scalp Electrode 1 ]       [ Scalp Electrode 2 ]
                 \                       /
                  \                     /
                   \                   /
                    [ Single Brain Source ]
             (Instantaneous, artificial synchronization)

Because this radiation happens at the speed of light, it creates an instantaneous, zero-lag correlation between those electrodes. If you calculate PLV here, you will get a deceptively high value, suggesting robust "functional connectivity" between two scalp regions when, in reality, you are just measuring a single brain region talking to itself through two different sensors.

To solve this, researchers use the Phase Lag Index (PLI). True neural communication between distinct brain regions is not instantaneous; it takes time for an action potential to travel down an axon and cross a synapse, resulting in a physical propagation delay or a phase lag.

PLI isolates this true communication by discarding all zero-lag synchronization. It looks at the sign of the phase differences:

$$\text{PLI} = \left| \frac{1}{N} \sum_{t=1}^{N} \text{sign}(\Delta\theta(t)) \right|$$

If the phase differences are clustered strictly around $0$ or $\pi$ radians (indicative of instantaneous volume conduction or perfect anti-phase reflection), the signs will balance out or be explicitly ignored, dropping the PLI to $0$. If there is a consistent, non-zero time lag (one region consistently leading or lagging the other), the signs will be highly asymmetric, yielding a high PLI.

By filtering out the physical "blurring" of electrical fields, PLI allows researchers to map true underlying neural interaction patterns during complex cognitive tasks (Teng Zhang et al.) and track subtle network reconfigurations within the brain's Default Mode Network (Muhammad Samran Navid et al.).

From Pairwise Synchrony to Graph Topologies

Once you can reliably calculate phase synchronization between every pair of sensors or reconstructed cortical sources, you can organize these relationships into a pairwise connectivity matrix. This matrix serves as the adjacency matrix for a brain graph, where nodes represent brain regions and edges represent the strength of their phase synchronization.

Pairwise Signals ----> Phase Sync (PLV/PLI) ----> Connectivity Matrix ----> Brain Graph
  [Ch 1] \                                         [1][2][3]                  (1)---(2)
          |=======>      Matrix Entry             [1] - .8 .2                  \   /
  [Ch 2] /                                        [2] .8 - .5                   (3)
                                                  [3] .2 .5 -

Transforming raw time-series data into abstract graph topologies allows researchers to leverage graph theory and deep learning to detect systemic changes in brain architecture.

For example, Abdullah Almohammadi demonstrated that feeding these phase-synchronized "brain graphs" into Graph Convolutional Neural Networks (GCNNs) yields state-of-the-art accuracy for motor imagery intent recognition, because the graph structure preserves the functional communication pathways irrespective of physical electrode distance.

Furthermore, these network representations reveal macro-level properties, such as "small-world efficiency"—the optimal balance between localized processing clusters and long-range integrative pathways. Chongyuan Lian et al. showed that during lexical retrieval tasks, patients with Mild Cognitive Impairment (MCI) exhibit a subtle breakdown in these global small-world properties in the theta band, long before traditional behavioral or amplitude measures show any noticeable decline. To capture these dynamics even more comprehensively, Manman Yuan et al. introduced hypergraph networks that fuse spectral phase synchronization with temporal connectivity, mapping multi-scale interactions that standard pairwise correlations miss.

Clinical and Real-Time Engineering Applications

The ability to extract clean, amplitude-independent signatures of neural coordination has driven phase synchronization from theoretical neuroscience into clinical engineering.

  • Closed-Loop Neuromodulation: Traditional deep brain stimulation (DBS) implants fire continuously or trigger based on simple local signal amplitude. Uisub Shin et al. designed a low-power System-on-Chip (SoC) capable of calculating inter-regional PLV in real time. This allows the implant to act as an intelligent gatekeeper, delivering therapeutic electrical stimulation only when pathological phase synchronization is detected between distinct brain regions, drastically reducing power consumption.
  • Objective Diagnostics: In critical care settings, distinguishing between states of unawareness is notoriously difficult. Adrien Witon et al. demonstrated that phase synchrony indices in the theta and alpha bands can objectively differentiate between patients in a coma and those in a quasi-brain-death state, providing a non-invasive diagnostic metric based on preserved network integration.
  • Adaptive BCIs for Rehabilitation: In motor imagery BCIs designed for stroke or neuromuscular rehabilitation, traditional spatial patterns often fail because the patient's underlying neural pathways have been damaged or reorganized. Kostas Georgiadis et al. showed that by exploiting the heightened, reorganized phase synchrony patterns inherent to patients with neuromuscular diseases, machine learning models can build more stable and reliable control interfaces than those relying on local signal magnitudes.

Ultimately, phase synchronization works because it aligns with the brain's fundamental mechanism for routing information. By treating the brain as a dynamic, clock-synchronized network rather than a collection of isolated power generators, this paradigm provides a clearer window into both healthy cognition and pathological disruption.

Key Foundational Work

Using Feedforward Neural Networks to Monitor Alertness from Changes in EEG Correlation and Coherence (1995) — contributed 2 concept elements to this direction.
Phase Synchrony Rate for the Recognition of Motor Imagery in Brain-Computer Interface (2005) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_phase_based_functional_connectivity_ecbd5878,
  title = {phase-based functional connectivity},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/phase-based-functional-connectivity/rdr_ecbd58788be141f4bb685407371088db},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_phase_based_functional_connectivity_ecbd5878_v1,
  title = {phase-based functional connectivity},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/phase-based-functional-connectivity/rdr_ecbd58788be141f4bb685407371088db/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake