When recording brain activity using non-invasive techniques like electroencephalography (EEG) or magnetoencephalography (MEG), sensors placed on the scalp pick up electrical potentials generated by populations of neurons in the cortex. A major goal in modern neuroscience is to map "functional connectivity"—how these different brain regions communicate and coordinate with one another.
However, a fundamental physical hurdle stands in the way: volume conduction. Because the brain, cerebrospinal fluid, and skull act as a conductive medium, electrical currents from a single neural source radiate outward and hit multiple surface sensors simultaneously. If you measure the correlation or synchronization between two sensors, they will appear tightly coupled. But this isn't because the underlying brain regions are communicating; it is simply because both sensors are listening to the exact same source. This signal leakage creates an illusion of connectivity, leading to widespread false positives in network mapping.
To solve this, researchers have developed a family of connectivity metrics that exploit a simple truth about physics and biology: volume conduction is instantaneous, but thinking takes time.
Exploiting Time Delays to Isolate True Communication
Volume conduction is an electromagnetic propagation phenomenon. At the scale of neural recording sampling rates, it happens instantaneously. Therefore, any artificial synchronization caused by signal leakage manifests with a zero-phase lag between sensors.
In contrast, true functional communication between distinct brain regions is biological. It requires neurons to fire action potentials, neurotransmitters to cross synapses, and signals to travel along axonal pathways. This process takes milliseconds. Consequently, when one brain region genuinely drives or synchronizes with another, a physical time delay occurs, resulting in a non-zero phase lag between their oscillatory signals.
By designing mathematical metrics that explicitly ignore or discount zero-lag interactions, researchers can filter out volume conduction artifacts and isolate true physiological coordination. As reviewed by Seraj in a comprehensive tutorial on cerebral synchrony (Cerebral Synchrony Assessment Tutorial), excluding zero-lag components allows researchers to separate true physiological coordination from the instantaneous signal spread inherent in scalp recordings.
The Phase Lag Index
The most direct way to operationalize this insight is the Phase Lag Index (PLI). Instead of measuring the absolute strength of phase coupling, PLI measures the asymmetry of the distribution of phase differences between two signals.
If we calculate the phase difference $\Delta \phi(t)$ between two sensors across many time points, we can look at whether one signal consistently leads or lags the other. If the synchronization is purely an artifact of volume conduction, the phase differences will cluster tightly around zero. Even with a bit of background noise, the phase differences will be symmetrically distributed around zero, meaning they are just as likely to be slightly positive as they are to be slightly negative.
Mathematically, PLI calculates the absolute average sign of the phase differences:
$$\text{PLI} = | \langle \text{sign}(\Delta \phi(t)) \rangle |$$
If the phase differences are centered at zero or evenly balanced, the positive and negative signs cancel out, resulting in a PLI of zero. But if true, time-delayed communication is occurring, one region will consistently lead the other, causing the phase differences to skew toward one side of the circle. The signs will not cancel out, and the PLI will approach one.
As demonstrated by Mao et al. in Frontotemporal phase lag index correlates with seizure severity in patients with temporal lobe epilepsy and Oguri in Phase Lag Analysis Scalp Electroencephalography May Predict Seizure Frequencies in Patients with Childhood Epilepsy with Centrotemporal Spikes, this phase-lag-based approach successfully quantifies functional connectivity in clinical conditions like epilepsy while preventing volume conduction artifacts. Similarly, Li et al. utilized this asymmetry of phase distributions in Source localization and functional network analysis in emotion cognitive reappraisal with EEG-fMRI integration to filter out zero-lag artifacts during emotion regulation tasks.
The Weighted Phase Lag Index
While the standard PLI is highly effective at removing volume conduction, it suffers from a major vulnerability: sensitivity to noise.
If two brain regions are weakly coupled and their true phase difference is very close to zero, small amounts of instrumental or environmental noise can cause the calculated phase difference to randomly flip signs back and forth (a phenomenon known as phase-slipping). Because PLI only cares about the sign of the difference and not its magnitude, these tiny, noise-driven fluctuations near zero can drastically distort the final connectivity score.
To solve this, researchers introduced the Weighted Phase Lag Index (wPLI). Instead of treating all non-zero phase differences equally, wPLI weights each phase difference by the magnitude of its imaginary component (which is proportional to the sine of the phase difference).
Phase differences that are very close to zero or $\pi$ have a tiny imaginary component, meaning their contribution to the overall score is heavily discounted. Phase differences that are closer to $\pi/2$ or $-\pi/2$ (representing clear, unambiguous lags) have a large imaginary component and are heavily weighted. This weighting scheme makes the metric exceptionally robust to noise and phase-slipping artifacts, as noted by Seraj (Cerebral Synchrony Assessment Tutorial).
This robustness explains why wPLI has become a gold standard for mapping functional networks in complex clinical populations where data can be noisy. For instance, Chatham et al. (Resting state EEG power spectrum and functional connectivity in autism), Sun et al. (Identification of diagnostic markers for ASD), and Jizi (Genetic modulation of brain dynamics in neurodevelopmental disorders) leveraged wPLI to map neural synchronization in autism spectrum disorders and neurodevelopmental conditions, ensuring that the observed network alterations reflect true regional interactions rather than signal crosstalk.
Applications in Network Reorganization and Cognitive Gating
Because phase-lag metrics isolate true physiological communication, they allow researchers to track how the brain dynamically reorganizes its networks during tasks or recovery.
In neurorehabilitation, Tang et al. (Bilateral upper limb robot-assisted rehabilitation improves upper limb motor function in stroke patients) used wPLI to show how robot-assisted training enhances interhemispheric motor network synchronization in stroke patients. Similarly, Sun et al. (Effects of visually induced motor imagery-based brain-computer interface training on motor function in patients with incomplete spinal cord injury) used wPLI to quantify network reorganization during spinal cord injury rehabilitation.
In cognitive neuroscience, Li et al. (The impact of cue and preparation prompts on attention guidance in goal-directed tasks) applied wPLI to map electrode interactions during goal-directed attention tasks, while Bertoni et al. (Pre-movement sensorimotor oscillations shape the sense of agency by gating cortical connectivity) demonstrated that the specific phase of pre-movement oscillations acts as a neural gate, orchestrating long-range cortical connectivity measured via zero-lag-filtered synchronization.
Furthermore, the utility of filtering out zero-lag artifacts extends to advanced clinical diagnostics, such as localizing seizure onset zones (Fang et al. in Epilepsy lesion localization method based on brain function network) and analyzing spike-free resting states in childhood epilepsy (Choi et al. in Electroencephalographic Resting-State Functional Connectivity of Benign Epilepsy with Centrotemporal Spikes).
Alternative Formulations of Time-Delayed Connectivity
The core philosophy—that true connectivity requires a time delay—is shared by other advanced metrics in the field.
For example, the Corrected Imaginary Phase Locking Value (ciPLV) similarly isolates the imaginary component of the cross-spectral density to eliminate zero-lag signal leakage. Sampaio et al. utilized this approach in Predicting Comatose Patient's Outcome Using Brain Functional Connectivity with a Random Forest Model to ensure that the network features used to predict patient recovery were not corrupted by volume conduction.
Taking this a step further, information-theoretic measures like Transfer Entropy (TE) look for directed, predictive relationships over time rather than simple phase synchronization. As Vicente et al. explain in Using transfer entropy to measure the patterns of information flow though cortex, because TE requires a time-delayed predictive relationship to establish effective connectivity, simultaneous signal leakage from volume conduction cannot satisfy its mathematical criteria. This makes information-theoretic predictive measures inherently resistant to volume conduction artifacts by design.
Key Foundational Work
Identifying brain effective connectivity patterns from EEG: performance of Granger Causality, DTF, PDC and PSI on simulated data (2011) — contributed 2 concept elements to this direction.
Using transfer entropy to measure the patterns of information flow though cortex: application to MEG recordings from a visual Simon task (2009) — contributed 1 concept elements to this direction.