DS-Qi: Quantitative EEG Biomarker in Pediatric Epilepsy

• DS-Qi is a quantitative EEG index defined as the sum of the posterior theta/alpha ratio and a beta-band desynchronization term, capturing local and global neural abnormalities.
• The index is derived from rigorously preprocessed, artifact-free scalp EEG recordings using standard 10-20 montage and validated across pediatric datasets.
• DS-Qi shows robust performance with significant correlation to seizure burden and high test-retest reliability, making it a promising tool for clinical outcome tracking.

DS-Qi is a two-feature quantitative EEG index proposed as an objective scalar biomarker of encephalopathic disease burden in pediatric epilepsy, with an initial focus on Dravet syndrome and external pre-validation on open pediatric EEG data. It is defined as the sum of a posterior theta/alpha power ratio and a beta-band desynchronization term based on the debiased weighted Phase-Lag Index, so that higher values reflect both greater posterior slowing and greater loss of beta-band synchrony. In the reported study, DS-Qi was derived from short, artifact-free scalp EEG segments recorded with a standard 10-20 montage and was presented as a reproducible single-number summary of electrophysiological severity rather than merely a binary detector (Uppalapati et al., 19 Sep 2025).

1. Definition and conceptual basis

The index is written as

$\mathrm{DS\text{-}Qi}_i = R_i + (1 - C_i),$

where $R_i$ is the posterior theta/alpha ratio for subject $i$, and $C_i$ is the global beta-band weighted Phase-Lag Index averaged across channel pairs. The additive structure aligns the directionality of the two constituents: elevated posterior slowing increases $R_i$, and reduced synchrony decreases $C_i$, so $1-C_i$ increases with greater network disconnection (Uppalapati et al., 19 Sep 2025).

The first component, $R_i$, targets excess posterior slow-wave power. In the study, this was motivated as a marker of impaired cortical maturation and as a phenomenon widely reported in pediatric epilepsy and various encephalopathies. The second component, $C_i$, targets beta-band synchrony. The use of the debiased weighted Phase-Lag Index was justified by its robustness to volume conduction and small-sample bias, and the transformation to $1-C_i$ was introduced so that both components increase with disease severity. This suggests that DS-Qi is explicitly designed to combine a local spectral abnormality and a global network abnormality within a single scalar summary.

2. Mathematical specification and feature extraction

The posterior slowing term is defined over the posterior electrodes O1, O2, P3, and P4 as

$R_i$0

where $R_i$1 and $R_i$2 are theta- and alpha-band powers on channel $R_i$3. In the reported implementation, theta was defined as 4–7 Hz and alpha as 8–12 Hz. Power spectral density estimation used Welch’s method with 2 s Hamming windows and 50% overlap (Uppalapati et al., 19 Sep 2025).

The synchrony term is computed in the beta band, defined as 13–25 Hz, across all unique pairs of retained channels:

$R_i$4

with $R_i$5 the number of retained channels. Here $R_i$6 denotes the debiased weighted Phase-Lag Index for channel pair $R_i$7 in the beta band. Averaging over all unique pairs produces a global connectivity measure rather than a region-specific graph statistic. Because the final index is

$R_i$8

a higher score can arise from either or both of two distinct electrophysiological patterns: increased posterior theta relative to alpha, and reduced beta-band phase consistency across channels.

3. Data sources and preprocessing pipeline

The external pre-validation used the CHB-MIT Scalp EEG Database as the epilepsy cohort and the EEG Motor Movement/Imagery dataset as the control cohort. The CHB-MIT set contributed 22 pediatric subjects aged 1.5–19 years, recorded at 256 Hz with 18–23 10-20 scalp electrodes. Controls were age-matched, with mean age $R_i$9 years, and were resampled to 256 Hz for direct comparison. For each subject, the analysis used a single 30 min awake, eyes-open, inter-ictal segment at least 10 min after any seizure, with the stated aim of avoiding post-ictal effects (Uppalapati et al., 19 Sep 2025).

The preprocessing chain consisted of bandpass filtering from 1–70 Hz, a 60 Hz notch filter, visual inspection to exclude drowsiness and artifact, independent component analysis for ocular and muscle artifact removal, and rejection of bad channels that were flatlined or excessively noisy. After cleaning, typically about 6 min of artifact-free EEG per subject remained. Controls were matched in age, vigilance state, and sampling rate. The paper emphasized strict rejection because electromyographic contamination can artificially increase beta connectivity and thereby bias the connectivity component of the index. A plausible implication is that the repeatability claims for DS-Qi depend not only on the feature design but also on rigorous artifact control, especially for the beta-band wPLI term.

4. Statistical validation and quantitative performance

In the reported comparison, the posterior theta/alpha ratio was $i$0 in epilepsy and $i$1 in controls, while beta wPLI was $i$2 in epilepsy and $i$3 in controls; both differences had $i$4. The composite DS-Qi was $i$5 in epilepsy versus $i$6 in age-matched normative EEG, with Cohen’s $i$7 and $i$8 (Uppalapati et al., 19 Sep 2025).

A logistic regression model using DS-Qi alone and evaluated with 10 × 10-fold cross-validation yielded an AUC of 0.90 with 95% CI 0.81–0.97. The optimal threshold by Youden’s index was DS-Qi = 1.46, at which sensitivity was 86% and specificity was 83%. The statistical workflow also included Shapiro-Wilk testing for normality, independent-samples $i$9-tests with Welch’s correction if needed or Mann-Whitney $C_i$0, ROC/AUC pooling, bootstrapping for confidence intervals, and Spearman correlation analysis. These details place DS-Qi within a conventional biomarker-validation framework rather than a purely descriptive EEG feature analysis.

Repeatability and severity association were also quantified. Across multi-day recordings, test-retest reliability was ICC = 0.74, with the detailed summary reporting $C_i$1 for 10 subjects. Higher DS-Qi correlated with greater seizure burden, with Spearman’s $C_i$2 and $C_i$3. Children in the upper seizure tertile $C_i$4 had DS-Qi $C_i$5, whereas the lower tertile $C_i$6 had DS-Qi $C_i$7; the reported contrast was $C_i$8, $C_i$9.

5. Interpretation, intended use, and scope of inference

The study positioned DS-Qi as a severity-oriented EEG biomarker rather than only a detector of epilepsy status. Its rationale is that it integrates two physiologically distinct axes: posterior slowing, interpreted as excess slow-wave activity and reduced cortical maturation, and beta-band desynchronization, interpreted as impaired cortico-cortical communication and network disruption. Because the output is a scalar, it is intended to be easy to track longitudinally and suitable for use as a secondary clinical-trial outcome, not just for binary classification (Uppalapati et al., 19 Sep 2025).

Operationally, the index was designed to be portable. The spectral component depends only on four posterior electrodes, while the synchrony component uses retained channels from a standard 10-20 recording. The authors emphasized that DS-Qi can be computed from short scalp EEG segments and described it as appropriate for retrospective data, prospective point-of-care assessment, and multi-center studies. The reported repeatability supports use in serial measurements across days, and the correlation with seizure count supports the interpretation of DS-Qi as an index of disease burden rather than a static group discriminator.

A frequent misconception would be to read the acronym as implying established Dravet-specific validation. The paper does not support that interpretation. The name reflects an initial focus on Dravet syndrome, but the reported evidence is an external pre-validation in broader pediatric epilepsy datasets. The title’s reference to a “roadmap to Dravet cohorts” is therefore substantive: it marks intended future cohort specificity rather than already completed syndrome-specific validation.

6. Limitations and future development

Three limitations were stated explicitly. First, there were no genetically confirmed Dravet cases in the present study, so the current evidence base is broader pediatric epilepsy rather than syndrome-specific validation. Second, the assessment was performed after medication withdrawal, which the paper notes may overestimate severity. Third, normative comparisons were limited to available open datasets (Uppalapati et al., 19 Sep 2025).

These limitations delimit the current interpretive scope of DS-Qi. The existing results establish discriminative performance, repeatability, and correlation with seizure burden in public pediatric EEG repositories, but they do not yet establish calibration for genetically defined Dravet cohorts, treatment-response sensitivity in routine clinical conditions, or invariance across broader normative reference populations. The paper’s stated trajectory is therefore not merely replication but cohort-specific extension: DS-Qi is presented as a compact and reproducible electrophysiological index whose next stage is prospective validation in Dravet cohorts.