Abstract
Background
Epilepsy possesses significant long-term health sequelae and potential morbidity among children globally. A diagnosis of seizures can often be challenging in the paediatric population owing to seizure mimics and absence of classical features in clinical as well as electroencephalographic (EEG) presentations.
Purpose
The objective of the study is to explore potential EEG biomarkers using quantitative analysis of conventional EEG records of subjects with seizure in paediatric populations.
Method
66 subjects with generalised seizures in the age group of 2–15 years had been recruited along with an equal number of age-matched healthy control subjects. Readings obtained from a 32-channel EEG were recorded along with the demographic details of the subjects. Using the Fast Fourier Transform (FFT) in the MATLAB environment, quantitative spectral features were extracted and compared between the groups and within the group to explore EEG markers related to the generalised seizures in the study population.
Results
The case group showed significantly higher alpha power in both left (17.86 vs 15.06 µV 2 /Hz) and right (20.76 vs 16.12 µV 2 /Hz) hemispheres compared to the controls. The case group had lower beta power in the left hemisphere (33.34 vs 35.52 µV 2 /Hz) but no difference in the right hemisphere. Significant asymmetries were observed across all cortical regions, including the temporal region showing the highest theta and delta asymmetry (asymmetry indices of −0.142 for both), whereas alpha asymmetry was highest in the occipitotemporal regions (asymmetry indices −0.125 and −0.141, respectively).
Conclusion
The study reveals a complex pattern of characteristic quantitative electroencephalographic (qEEG) alterations in the paediatric seizure population suggestive of significant neurophysiological implications, including alterations in arousal, attention and inhibitory control mechanisms and the features may be of some value in objective seizure assessment using EEG.
Introduction
Epilepsy is the second most common neurological disorder globally, posing significant health and social burdens for patients, families and healthcare systems. Around 75% of epilepsy cases begin during childhood, highlighting the increased vulnerability of the developing brain to seizures. Although prevalence rates vary—with estimates in India ranging from 1.3 to 11.9 per 1,000 people—the global burden remains high, affecting nearly 70 million individuals, most of whom live in developing countries. 1 In children, seizures are associated with several complications, including cognitive delays, 2 behavioural disturbances, 3 adverse effects of long-term antiepileptic drug use and an increased risk of injury or death. Common contributing factors include febrile illnesses, birth-related brain injuries, infections and congenital brain malformations.
Although seizure disorders have been recognised since ancient times, diagnosis still heavily relies on clinical presentation due to the intricate nature of the condition. While significant strides have been made in understanding its pathophysiology, diagnosis, particularly in paediatric cases, remains complex. Electroencephalography (EEG) stands as a cornerstone in diagnosis, offering insights into brain electrical activity. However, its sensitivity ranges from 35% to 56%, and specificity varies from 78% to 98%, highlighting the need for improved diagnostic tools, especially in the paediatric population where seizure mimics are prevalent, and certain seizure types lack classical features.4–6
Addressing the diagnostic challenges, quantitative electroencephalographic (qEEG) analysis emerges as a promising avenue. By examining mathematical features such as power spectral density (PSD), qEEG aims to quantify voltage and time relationships in EEG recordings, providing objective insights into neuronal activities. Traditionally, frequency parameters have been pivotal, with attention drawn to specific frequency bands like delta, theta, alpha, beta and gamma. Recent advancements in computer-based EEG recording and analysis have enhanced the understanding and utilisation of qEEG, offering a more comprehensive approach to seizure detection and prediction.7–9
Key to qEEG analysis is the identification of seizure-affected locations within EEG signals. Manual inspection proves impractical given the small proportion seizures occupy within the total signal. Hence, the focus shifts towards quantitative assessment during interictal periods, moving beyond classical spike-and-wave features. By analysing power distribution across frequency bands, qEEG can unveil subtle abnormalities indicative of seizure activity, even in conventional EEG records that appear normal.
In the present study, qEEG features were compared between cases and age-matched controls using PSD analysis to understand the EEG-related objective biomarkers within the EEG signals. The aim was to discern qEEG biomarkers specific to different frequency bands, aiding in the identification of seizure patterns within seemingly normal EEG recordings of paediatric patients.
Methods
Participants
This observational case-control study enrolled participants aged 2 to 15 years to investigate seizure-related characteristics. The case group included drug-naïve children with generalised seizures who sought care at the paediatric outpatient and inpatient departments (OPD/IPD) of AIIMS Patna, India, and had experienced at least one seizure episode in the preceding two months. The control group consisted of age-matched children from the same hospital, without a history of seizures, who attended for non-neurological reasons, such as immunisation.
Exclusion criteria for both groups included children with severe comorbidities such as history of head injuries, status epilepticus, metabolic encephalopathy or other neurological disorders known to impact EEG findings. Additionally, children with focal seizures, hydrocephalus or cranial malformations were also excluded from the study.
Sample size calculation was based on the Taro Yamane formula 10 for hospital-based observational studies. With an average of 200 paediatric patients referred for EEG due to seizures over the past two to three years, the total number of potential cases (N) was estimated at 200. Using a margin of error (e) of 10%, the sample size (n) was calculated to be 66. The study utilised a 1:1 case-control ratio, resulting in a control group size of 66. Consecutive sampling was employed, where participants were sequentially recruited as they became available, ensuring a systematic and unbiased selection process. Data collection occurred over an 18-month period, from September 2021 to March 2023.
The study received ethical approval from the Institute Ethical Committee (IEC No. AIIMS/Pat/IEC/PGTh/Jan 21/27 dated 29 December 2021). Informed consent was obtained from the parents or guardians of all eligible participants after clear explanation of the study procedures in both Hindi and English. Written informed consent was secured from the guardians of all participants. Additionally, verbal assent was obtained from children aged 7–12 years and written assent from children aged 12–15 years, in accordance with standard ethical guidelines.
Procedure
Demographic Variables
Demographic information was collected from the attendants of the subjects using a detailed demographic questionnaire. The information collected includes detailed antenatal and natal history, including type of delivery and history of birth asphyxia, family history, details of immunisation, birth weight, etc. The questionnaire also included details of seizure history, its frequency and type.
EEG Data Acquisition
Continuous EEG recordings were obtained from 66 control and 66 case paediatric participants, each for a duration of 20 minutes, while in a relaxed state with eyes closed to minimise artefacts from visual input and attention. All recordings were conducted during the daytime between 9:00 am and 5:00 pm, in accordance with the standard protocols of the neurodiagnostic facility at the institute. Prior to EEG acquisition, the patients underwent preparation, which included scalp cleansing, marking electrode positions, the application of a viscous conductive electrolyte gel and the fitting of the EEG cap.
EEG signals were recorded using a 32-channel actiCHamp Plus EEG amplifier system (Brain Products GmbH, Germany) at a sampling frequency of 500 Hz. The recordings took place in a semi-darkened, electromagnetically shielded room within the neurophysiology laboratory of the institute. Subjects were seated upright during the procedure. Of the 32 channels, 19 were configured in a unipolar montage (C3, C4, F3, F4, F7, F8, Fz, Fp1, Fp2, O1, O2, Oz, P3, P4, P7, P8, Pz, T7 and T8), while the remaining channels were configured in a bipolar montage. The unipolar channels, covering all five brain regions, were the primary focus of the analysis.
Data Pre-processing 11
The EEG data pre-processing steps included filtering, artefact removal and the data segmentation process. The details of the signal processing pipeline have been summarised in Figure 1.
Block Diagram of the Proposed Methodology of Generalised Seizure Diagnosis in Paediatric Patients Using EEG Signals.
The signals underwent filtration through tenth-order digital Butterworth bandpass filter having cut-off frequencies of 0.3 Hz and 70 Hz, which effectively removed undesired high-frequency components, slow drifts and DC components. To filter out the noise components resulting from power line interference, a fifth-order digital notch filter with a stopband of 50 Hz was utilised to pre-process the EEG signals. After this, an independent component analysis (ICA) was applied to eliminate the artefacts caused by muscle activity, eye blinks and eye movements. These steps were intended to improve data quality and enhance the reliability of EEG recordings for subsequent analysis and interpretation. To facilitate analysis and reduce computational load, the data had been segmented into smaller, manageable time windows of 30 seconds artefacts-free EEG epochs. A sample waveform of pre-processed EEG from a healthy control and that of a subject with seizures is depicted in Figure 2.
Waveform of Single Channel EEG Data of: (a) Healthy Control (HC), and (b) Epileptic Seizure Symptomatic Paediatric Patient.
Signal Decomposition into EEG Rhythms
The EEG signals were further subjected to filtration through a filter bank, leading to the decomposition of EEG signals into four distinctive EEG rhythms, that is, delta (δ) ranging from 0.5 Hz to 3.9 Hz, theta (θ) ranging from 4 Hz to 7.9 Hz, alpha (α) ranging from 8 Hz to 12.9 Hz and beta (β) ranging from 13 Hz to 30 Hz.
Each filter in the bank is designed optimally using the Parks-McClellan (PM) algorithm, with passband and stopband cut-off frequencies determined based on EEG signal spectral characteristics. 12 The provision of the sampling frequency is essential for determining the optimal order of FIR filter using PM algorithm. A bandpass filter bank, comprising Equiripple optimal FIR filters, was employed to decompose pre-processed EEG signals into four distinct EEG bands. Additionally, each band was thoroughly examined to identify potential epileptic seizure biomarkers. Furthermore, the digitised EEG data underwent processing and analysis within a MATLAB (Version: 9.13.0; R2022b) environment.
Feature Extraction
In the context of qEEG, the PSD feature is one of the important measures used to analyse and interpret brainwave activity. This average band power or PSD is typically estimated using the FFT algorithm of discrete signals. The average band power or PSD in each frequency band have been normalised to derive the relative PSD of each band in relation to the entire frequency spectrum. The PSD provides valuable insights into the functioning of the brain and can be used to examine various cognitive processes, emotional states and neurological disorders. In this study, each EEG epoch was analysed to extract the prominent PSD-based relative power feature. The occurrence of a seizure can be predicted by the nature of the PSD feature.
Data Division
To identify potential EEG-based biomarkers for paediatric epileptic seizures, three distinct data sets were derived from the original EEG recordings using a structured data division approach. These variations allowed for targeted analysis of the impact of different frequency bands, anatomical brain regions and hemispheric activity on seizure diagnosis.
In the first data set, frequency-specific EEG components obtained after signal decomposition were analysed band-wise for both epileptic and control groups. This enabled assessment of the diagnostic relevance of individual brain rhythms.
In the second data set, the spatial distribution of EEG activity was considered. The 19 EEG channels were grouped according to their anatomical correlation with five major brain regions: frontal (F3, F4, F7, F8, Fp1, Fp2, Fz), central (C3, C4), parietal (P3, P4, P7, P8, Pz), occipital (O1, O2, Oz) and temporal (T7, T8). This classification aimed to evaluate region-specific biomarkers.
The third data set focused on hemispheric analysis by organising EEG channels into left and right hemispheres to study lateralisation patterns associated with epileptic activity.
This structured data segmentation was designed to systematically investigate how different EEG features—frequency bands, cortical regions and hemispheric localisation—contribute to the accurate diagnosis of paediatric epilepsy data analysis and interpretation
The data have been tested for normality using Q-Q plot in the statistical software SPSS (IBM SPSS Statistics software Ver. 23, USA). All the normally distributed data has been expressed as mean and standard deviation and analysed using parametric tests, including independent sample t-tests. The non-normally distributed data have been expressed as median and mode with interquartile range and were analysed using non-parametric tests, including the Mann-Whitney U test for independent samples. The independent variables included gender, seizure frequency and seizure duration, while outcome variables included EEG parameters (PSD alpha, PSD beta, PSD theta, PSD delta, etc.). The p value of .05 with 95% confidence limit has been considered as a level of statistical significance for the study.
Result
Table 1 presents the sociodemographic characteristics, antenatal, natal and postnatal history of both cases and controls, with age and gender-matched participants. All the demographic parameters showed uniform distribution across the cases and the control population.
Demographic Parameters of Case and Control Group Participants.
The overall qEEG PSD analysis shows a statistically significant distribution asymmetry in the spectrum of alpha, theta and delta band PSD between the control and study group participants, with the study group showing significantly higher mean PSD in comparison to the control group participants (see Table 2).
PSD in Various Frequency Ranges of Overall Brain Cortical Area Among the Study and Control Group Participants.
Further analysis of lobe-wise distribution of the PSD in various frequency spectrums (see Table 3) revealed that the PSD of beta activity showed significantly lower power distribution on the left side among study participants in comparison to the control population. In all other frequency spectrums, the PSD was higher than that of the control population on both sides. Unlike the left side, the right side beta activity PSD did not show any significant difference from that of control group participants.
Mean PSD of Various Frequency Bands in Left and Right Cortical Regions of Study and Control Group Participants.
Lt = Left; Rt = Right.
To evaluate the possible interhemispheric asymmetry in the distribution of the PSD, the asymmetry index 13 had been computed by the equation:
Asymmetry Index = (Right Hemisphere PSD − Left Hemisphere PSD) / (Right Hemisphere PSD + Left Hemisphere PSD).
The negative values of the asymmetry index indicate a leftward asymmetry, whereas positive values indicate a rightward asymmetry (Table 4). Except for the beta frequency spectrum, all the other frequency spectrums have shown interhemispheric asymmetry among the study participants. The frontal region presented with rightward asymmetry in the alpha, theta and delta bands. The theta and delta bands also showed significant rightward asymmetry. All the other areas had shown leftward asymmetry.
Left-right Asymmetry in PSD Across Various EEG Frequency Bands in Different Cortical Regions Among Study Participants.
Discussion
The interictal qEEG has a significant role as a biomarker for seizure diagnosis. The PSD, especially in the slower spectrum, has a significant contribution to this, as well. 14 The presented data reveal several interesting findings regarding EEG PSD differences between the case and control groups, as well as hemispheric asymmetries within the case group. The case group showed significantly higher overall PSD in the alpha, delta and theta bands compared to controls (Table 2). This suggests a general increase in low-frequency oscillatory activity across the cortex in the case group. The increased alpha PSD (19.86 vs 15.7 µV2/Hz) may indicate enhanced cortical idling or inhibition in the case group. Alpha oscillations are thought to reflect top-down inhibitory control processes 15 and play a key role in attention and information gating. 16 The higher alpha power could suggest altered attentional processes or an increased internal focus in the case group. The elevated delta and theta power (6.23 vs 3.86 µV2/Hz for both) is particularly noteworthy. Increased low-frequency oscillations are often associated with cognitive impairment, drowsiness or pathological conditions. 17 Delta oscillations are linked to motivational processes 18 and theta to memory and emotional regulation. 19 The higher power in these bands may reflect alterations in arousal, emotional processing or basic cognitive functions in the case group. Table 3 reveals some intriguing lateralisation effects in the case group.
The case group showed significantly higher alpha power in both left (17.86 vs 15.06 µV2/Hz) and right (20.76 vs 16.12 µV2/Hz) hemispheres compared to the controls. Notably, the right hemisphere increase was more pronounced. This right-lateralised alpha increase could indicate greater right hemisphere involvement in attentional or inhibitory processes.
Interestingly, the case group had lower beta power in the left hemisphere (33.34 vs 35.52 µV2/Hz) but no difference in the right hemisphere. Beta oscillations are associated with active cognitive processing and motor control. The reduced left hemisphere beta may suggest altered cognitive or motor functions lateralised to the left side. 20
Both hemispheres showed increased delta and theta power in the case group, consistent with the overall findings. This bilateral increase in slow-wave activity could indicate widespread alterations in arousal or basic cognitive processes.
Significant asymmetries were observed across all cortical regions. Notably, the frontal region showed higher alpha in the left hemisphere, while the central, occipital, parietal and temporal regions showed higher alpha in the right hemisphere. This complex pattern of alpha asymmetry may reflect region-specific alterations in inhibitory control or attentional processes.
No significant beta asymmetries were observed, suggesting relatively balanced high-frequency activity across hemispheres.
Similar asymmetry patterns were seen for delta and theta bands. The frontal and central regions showed higher power in the left hemisphere, while occipital and temporal regions showed higher power in the right hemisphere. This asymmetric distribution of slow-wave activity could indicate lateralised alterations in arousal or basic cognitive functions.
These findings reveal a complex pattern of EEG alterations in the case group, characterised by a general increase in low-frequency oscillations, region-specific hemispheric asymmetries and some lateralised effects in higher frequency bands. The neurophysiological implications may include alterations in arousal, attention, inhibitory control and basic cognitive processes. 21 Further research is needed to elucidate the functional significance of these EEG patterns and their relationship to specific cognitive or clinical characteristics of the case group.
In the domain of paediatric neurology, understanding and accurately diagnosing seizure disorders pose significant challenges. The intricacies of these conditions require a computational approach that goes beyond conventional diagnostic methods, which rely on classical features like spike-and-waves. Through the integration of cutting-edge methodologies such as qEEG analysis and comprehensive clinical observations, the study aimed to offer significant advancements to the field.
Hemispheric Asymmetry in Various Power Bands
Left-right asymmetry in EEG PSD across various frequency bands has been well reported in patients with epilepsy, 22 highlighting the hemispheric differences in cortical excitability and seizure propagation pathways. Studies have shown that seizure-prone regions often exhibit asymmetric PSD in low-frequency bands, such as delta (0.5–4 Hz) and theta (4–8 Hz), particularly in focal epilepsy. Increased power in these slower bands is commonly observed ipsilaterally to the seizure focus due to disrupted neuronal networks and cortical dysfunction in the affected hemisphere. Moreover, asymmetry in fast frequencies, especially beta (13–30 Hz) and gamma (>30 Hz), has been associated with epileptogenic zones, reflecting abnormal hyperexcitability and synchrony of cortical neurons. The focal seizures are often known to be preceded by an increase in low-frequency PSD in the affected hemisphere, which correlates with impaired neuronal communication and cortical inhibition mechanisms.
This asymmetry also extends to interictal periods, where patients with temporal lobe epilepsy frequently show reduced high-frequency power (beta and gamma) in the seizure focus hemisphere compared to the contralateral side, 22 suggesting persistent cortical dysfunction outside of seizure events. Such asymmetries can have both diagnostic and prognostic implications. For instance, asymmetrical theta and alpha band PSD has been used as a biomarker for lateralising the seizure focus in pre-surgical evaluations. 23 Importantly, non-epileptic individuals generally exhibit more symmetrical PSD across hemispheres, further underscoring the pathological nature of these asymmetries in epilepsy. Recent advancements in quantitative EEG analysis, including machine learning approaches, have shown promise in leveraging these asymmetries to improve the accuracy of seizure localisation and prediction. Understanding the relationship between hemispheric asymmetry in PSD and seizure dynamics can enhance the precision of epilepsy management strategies, including surgical intervention planning.
Conclusion
The diagnosis of seizure disorder is challenging in the paediatric population owing to variations in presentation as well as the limited role of clinical EEG. qEEG has the potential to highlight certain patterns buried within normal-looking EEGs of children with a history of clinical seizures. The principle identifying features which may act as a biomarker for the seizure include the presence of interhemispheric asymmetry in the interictal short-term EEG records, primarily in the alpha, theta and delta bands of the frontal and central regions.
Footnotes
Acknowledgement
The work is an outcome of the thesis project undertaken in the Department of Physiology of AIIMS Patna. All the authors express a deep sense of gratitude to all the department faculty and associated lab staff for their cooperation. The support of the Institute is also dully acknowledged.
Authors’ Contribution
Arundhati Kumari has conceptualized the work, Kamlesh Jha and Tribhuwan Kumar has supervised the work and prepared the manuscript. Lokesh Tiwari has supervised the clinical subject recruitment. Rakesh Ranjan and Purushottam Kumar has taken care of statistical analysis of the collected data.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.
Funding
The authors received no financial support for the research, authorship and/or publication of this article.
Statement of Ethics
The study has received ethical approval from the Institutional Ethical Committee of AIIMS Patna (IEC No. AIIMS/Pat/IEC/PGTh/Jan 21/27 dated 29 December 2021. All the due ethical norms have been followed during the whole study procedure.
