Temporal-Parietal Phase Clustering via EEG

• Temporal-parietal phase clustering is defined as beta-band synchrony between temporal and parietal EEG sites, computed via Hilbert transform and phase-locking value.
• The methodology features robust EEG preprocessing, instantaneous phase extraction, and sliding window PLV calculations for real-time cognitive state monitoring.
• Empirical results show that TC–PC features yield up to 95% classification accuracy, underscoring its applicability in brain-computer interfaces under varying illumination.

Temporal-parietal phase clustering refers to β-band phase synchrony between temporal and parietal cortical sites, measured via EEG. As operationalized by Li et al. (Li et al., 19 Dec 2025), it provides a robust and low-dimensional electrophysiological marker for distinguishing working memory states and supports cognitive ergonomics assessment in brain-computer interface (BCI) applications. The metric is computed by extracting β-band instantaneous phase from temporal (FT9, T7) and parietal (P3, Pz) electrodes, calculating the phase-locking value (PLV) across these regions within sliding time windows, and analyzing both temporal dynamics and an explicit phase-space clustering structure.

1. Methodological Pipeline for Temporal–Parietal Phase Clustering

The methodology consists of the following steps:

EEG Acquisition and Preprocessing

• EEG data are recorded from 32 Ag/AgCl electrodes positioned according to the 10–20 system, sampled at 1,024 Hz and subsequently down-sampled to 128 Hz.
• Re-referencing utilizes a whole-head average reference.
• Zero-phase finite impulse response band-pass filtering (MATLAB/EEGLAB) is applied from 0.5–40 Hz.
• Artifact rejection combines ICA (runica) for ocular/muscle components, automatic epoch exclusion (exceeding ±100 μV), and visual inspection.
• For temporal-parietal (TP) analysis, FT9 and T7 serve as temporal channels, P3 and Pz as parietal channels.

Extraction of β-Band Phase Series

• The β-band is defined with lower and upper cut-offs at 13 Hz and 30 Hz, respectively.
• For each selected channel, the Hilbert transform yields the analytic signal and instantaneous phase:

$$x_H(t) = PV \int_{-\infty}^{\infty} \frac{x(\tau)}{t-\tau} d\tau$$

$\phi_x(t) = \arg[ x(t) + i \cdot x_H(t) ]$

PLV Computation

• For each pair of temporal (t) and parietal (p) channels, at each time sample $n$:

$\Delta\phi_{tp}(n) = \phi_t(n) - \phi_p(n)$

• Over a window of $N$ samples (128 samples = 1 s, 50% overlap), the phase-locking value is:

$$PLV_{tp} = \left| \frac{1}{N} \sum_{n=1}^{N} e^{i \cdot \Delta\phi_{tp}(n)} \right|$$

where $0 \leq PLV_{tp} \leq 1$.

• PLV values across both temporal and both parietal channels are averaged to produce a single $R_{TC–PC}(t)$ trace per subject.

Dynamic Phase Maps and Clustering

• $R_{TC–PC}(t)$ is computed in 1 s sliding windows stepped every 0.5 s.
• Additional curves representing prefrontal-temporal and prefrontal-parietal PLVs ($R_{TC–PFC}(t), R_{PC–PFC}(t)$) are constructed for comparative analysis.
• Three-dimensional phase-space maps are formulated as $[R_{TC–PC}(t), R_{PFC–TC}(t), R_{PFC–PC}(t)]$, with relevant two-dimensional projections.
• For each working memory (WM) state, cluster centroids are calculated by averaging $R(t)$ within the state.

Balance Metrics

Two measures quantify balance between temporal and parietal synchrony:

• Geometric distance (GD): For a centroid $(x_0, y_0)$ to the balance line $y = x$

$GD = \frac{|A x_0 + B y_0 + C|}{\sqrt{A^2 + B^2}}$

with $A=1$, $B=-1$, $C=0$.

• Arithmetic difference: $h(t) = R_{TC–PC}(t) - R_{TC–PC}(t)_{BalanceLine}$, zero-centered at perfect TP balance.

2. Empirical Findings: PLV and State Discrimination

The principal findings demonstrate that temporal–parietal β-band PLV robustly distinguishes among working memory states, modulated by environmental illumination:

State	300 lx Recall (S1)	300 lx Sequence (S2)	1000 lx Recall (S3)	1000 lx Sequence (S4)
Mean PLV (±SD)	0.62 ± 0.04	0.59 ± 0.05	0.52 ± 0.06	0.49 ± 0.05

• Paired t-test (S1 vs S4): $t(21)=8.1$, $p<0.001$, Cohen’s $d=1.57$.
• Increasing illumination yields decreased $PLV_{TC–PC}$ (temporal–parietal synchrony) and a compensatory increase in $R_{PC–PFC}$ (parietal–prefrontal synchrony), indicating dynamic rebalancing under cognitive load or environmental demands.

3. Quantitative Characterization and Statistical Testing

Balance metrics provide finer discrimination:

| Condition | GD mean (±SD) | Arithmetic mean(|h|) (±SD) | |-----------|---------------------|----------------------------| | S1 | 0.200 ± 0.032 | 0.17 ± 0.04 | | S2 | 0.183 ± 0.028 | | | S3 | 0.122 ± 0.025 | | | S4 | 0.081 ± 0.022 | 0.06 ± 0.02 |

• One-way ANOVA across states: $F(3,63) = 47.2$, $p<0.0001$. Post hoc S1 > S4, $p<0.001$.
• $|h|$ decreases significantly from S1 to S4 ($p<0.001$), consistent with TP balance being optimized as illumination increases.

4. Machine Learning Validation and Feature Dominance

Classification experiments confirm temporal-parietal β PLV ($R_{TC–PC}$) as the dominant feature for distinguishing WM states across tasks:

Features	Gradient Boosting	Random Forest	k-NN	Decision Tree
TC–PC	0.92 (0.95)	0.90 (0.93)	0.95 (0.98)	0.85 (0.89)
TC–PFC	0.84 (0.88)	0.82 (0.87)	0.86 (0.90)	0.78 (0.83)
PC–PFC	0.77 (0.81)	0.75 (0.80)	0.78 (0.82)	0.70 (0.76)
TC–PC–PFC	0.88 (0.92)	0.86 (0.90)	0.89 (0.94)	0.80 (0.85)

(Accuracy with AUC in parentheses; peak: k-NN 0.95 accuracy, 0.98 AUC). TC–PC features consistently outperform all alternatives, both as single-region PLV and in multivariate combinations.

5. Interpretation and Practical Recommendation for BCI Systems

Temporal-parietal phase clustering enables real-time cognitive ergonomics monitoring and resource-efficient BCI operation:

• $R_{TC–PC}(t)$ and $h(t)$ can be computed in 500 ms windows with 50% overlap, yielding a 2 Hz update rate—suitable for real-time ergonomics.
• A threshold $|h| < 0.05$ predicts an "optimized" memory state with $>$90% sensitivity and specificity.
• Implementation is feasible using only four EEG channels (FT9, T7, P3, Pz), reducing the system’s complexity and wire footprint.
• Recommended workflow: band-pass filtering (13–30 Hz), Hilbert transform for phase extraction, and PLV computation in overlapping windows; classification via $k$-NN ($k=5$) or thresholding $h(t)$ achieves 90–95% accuracy with latency $<$200 ms.
• The pipeline is directly transferable to platforms such as MATLAB, Python (MNE/NumPy/SciKit-Learn), or embedded BCI devices.

6. Environmental Illumination and Functional Implications

Alterations in ambient illumination modulate temporal-parietal synchrony, with higher illumination (1,000 lx) reducing $R_{TC–PC}$ and increasing parietal dominance. Beta-band power maps confirm heightened P3/Pz activity under these conditions, reflecting a neuroergonomic adaptation. The observed illumination-dependent rebalancing supports use of TP phase clustering as a noninvasive marker of optimized cognitive states in real-world environments.

7. Limitations and Perspectives

Findings from Li et al. are based on 22 subjects and specific working memory paradigms (Li et al., 19 Dec 2025). A plausible implication is that the robustness of temporal-parietal phase clustering may generalize to other cognitive contexts, but empirical validation is required. The methodology relies on high-quality artifact rejection; residual noise or differences in electrode placement could impact reproducibility. Further work is needed to examine potential sensitivity to inter-individual variability, task complexity, and external distractors, as well as to extend the pipeline to higher-density EEG arrays or other synchronization metrics.