Nigerian Schizophrenia EEG Dataset

• NSzED is a comprehensive EEG dataset capturing neural activity from Nigerian schizophrenia patients and healthy controls using standardized clinical assessments.
• It employs established EEG acquisition and preprocessing techniques, including baseline correction and band-pass filtering, to ensure high-quality signal integrity.
• The dataset supports advanced machine learning studies and connectivity analyses to develop population-specific diagnostic tools and reduce research biases.

The Nigerian Schizophrenia EEG Dataset (NSzED) is a structured repository of electroencephalographic recordings from clinically characterized schizophrenia patients and healthy controls originating from Nigeria. It is the first EEG schizophrenia dataset from West Africa and the African continent, assembled with a view toward advancing neuropsychiatric research and supporting the development of automated EEG-based diagnostic and prognostic tools tailored to underrepresented populations (Olateju et al., 2023). Below is a systematic and technical overview of the NSzED, contextualized within the framework of modern computational psychiatry and related EEG-based schizophrenia classification research.

1. Dataset Structure and Clinical Characterization

NSzED comprises EEG recordings from 37 schizophrenia patients and 22 healthy control subjects, all of Nigerian origin. Diagnostic categories are established using the Mini International Schizophrenia Interview (MINI), a structured clinical assessment tool. Symptom severity is quantified using the Positive and Negative Symptoms Scale (PANSS), and functional disability is measured via the World Health Organization Disability Assessment Schedule (WHODAS). All patients are in-patient populations at the Obafemi Awolowo University Teaching Hospital Complex (OAUTHC) or its subsidiary, Wesley Guild Hospital. Controls are predominantly local volunteers, including students and clinicians affiliated with the same medical centers. Rigorous inclusion and exclusion procedures are applied; subjects with substantial recording artifacts, concurrent substance abuse, or comorbid neurological disorders are excluded to ensure data integrity (Olateju et al., 2023).

2. EEG Acquisition Protocols and Experimental Conditions

Recordings utilize the international 10/20 electrode placement system, implementing 19 scalp electrodes (Fp1, Fp2, F3, F4, F7, F8, C3, C4, Cz, T3, T4, T5, T6, P3, P4, Pz, O1, O2), ensuring comprehensive spatial coverage of anterior frontal, frontal, central, temporal, parietal, and occipital regions. Two EEG devices are used:

• Contek-2400: 200 Hz sampling rate
• BrainAtlas Discovery-24E: 256 Hz sampling rate

Protocol consists of three cognitive/behavioral states, enabling broad neurofunctional interrogation:

• Restful state (baseline, eyes closed/eyes open)
• Mental arithmetic task execution: Subjects receive arithmetic problems visually and verbally, often in their native language, to induce cognitive effort.
• Auditory stimuli tasks:
  • Auditory oddball: Inclusion of standard (1kHz, 100ms) and deviant (1kHz, 200ms; 3kHz, 100ms) tones, passively attended, for event-related potential (ERP) analysis (e.g., mismatch negativity, MMN).
  • Auditory steady-state (ASSR): A 40Hz tone protocol (especially for BrainAtlas device) eliciting steady-state responses.

Synchronization methods integrate verbal/visual cues or a hand-held clicker, marking task onsets and responses in the EEG metadata for precise alignment between tasks and neural data (Olateju et al., 2023).

3. Data Formats, Organization, and Technical Processing

NSzED is structured into device-specific directories with sub-folders per subject/session. Core formats include:

• EDF (European Data Format): Standard time-series files of continuous EEG
• .sav files: Preprocessed data for immediate feature extraction
• GNR files/spreadsheets: Session metadata, including device ID, synchronization scheme, and clinical/demographic attributes

Raw data undergo several pre-processing operations:

• Baseline correction
• Band-pass filtering (1–100 Hz)
• Notch filtering at 50 Hz (to remove line noise)
• Segmentation into tasks/epochs, with exclusion of segments containing excessive artifacts

For ERP computation, epochs are created post-stimulus, especially for the MMN, targeting five consecutive latency windows ([0–100 ms], [100–200 ms], [200–300 ms], [300–400 ms], [400–450 ms]) to accommodate intersubject variability (Olateju et al., 2023).

4. Feature Computation: Biomarkers and Complexity Metrics

NSzED is designed for the extraction and analysis of multiple established EEG biomarkers:

• Mismatch Negativity (MMN): Computed as the difference between average ERP responses to standard and deviant tones over several post-stimulus windows, capturing the neural signature of aberrant auditory processing pervasive in schizophrenia.
• Fuzzy Entropy: Quantifies the irregularity and complexity of the EEG time series, computed by averaging fuzzy entropy across epochs and spatially grouped cortical regions. Increased/decreased entropy may index cortical disorganization (Olateju et al., 2023).
• Auditory Steady-State Response (ASSR): Extracted from 40Hz stimulation runs, using narrowband filtering and analytic amplitude estimation via Hilbert transform and FFT. ASSR abnormalities are linked to neural synchrony deficits.

Although these features are calculated as descriptive markers within the dataset documentation, the underlying raw and preprocessed data are suitable for higher-order feature extraction pipelines, including power spectral density, coherence, and connectivity metrics found in advanced classification studies (Olateju et al., 2023, Singh et al., 6 Feb 2025, Phang et al., 2019).

5. Relevance to Computational Psychiatry and Machine Learning Applications

NSzED furnishes a unique resource from an underrepresented population, enabling the development and validation of EEG-based diagnostic and prognostic models that avoid Eurocentric or Asian sampling bias. Its multi-paradigm experimental design supports analyses spanning:

• Supervised learning classification tasks (e.g., schizophrenia vs. control) with deep learning (CNN, RNN, autoencoder) and traditional ML (SVM, RF, decision tree) architectures (Chu et al., 2017, Wu et al., 2022, Saadatinia et al., 2023, Singh et al., 6 Feb 2025, Alkhalifa, 6 Mar 2025, Jin et al., 11 Aug 2025).
• Functional connectivity analysis (VAR, PDC, complex network measures), supporting connectomics approaches to neuropsychiatric biomarker discovery (Phang et al., 2019, Alves et al., 2021).
• Event-related and oscillatory marker assessment, e.g., MMN latency/area, ASSR amplitude, band-specific power/coherence, and entropic features (Olateju et al., 2023).

The dataset’s completeness and metadata facilitate direct integration into distributed processing (e.g., Apache Spark) or large-scale deep learning pipelines, supporting scalable research designs (Singh et al., 6 Feb 2025).

6. Technical Considerations and Data Quality

All recordings are validated by clinicians and electrophysiologists for signal integrity. Sessions with major artifacts, excessive noise, or confounding neurological/psychiatric histories (outside primary diagnosis) are excluded. The inclusion of multiple device types and task protocols necessitates careful normalization or domain adaptation in downstream ML analyses (e.g., channel matching, re-referencing, or transfer learning). Data are provided in formats compatible with both open-access tools and bespoke pipelines, permitting flexible research workflows (Olateju et al., 2023).

7. Scientific Significance, Accessibility, and Future Directions

NSzED represents a landmark in African clinical neuroinformatics, offering:

• The first, validated, multi-condition EEG schizophrenia dataset from the continent;
• Open, reproducible data formats with comprehensive demographic and clinical metadata;
• The capacity for cross-cultural validation and extension of machine learning models, potentially leading to population-specific biomarker discovery and diagnostic applications;
• Support for future additions, including multimodal data and extended phenotyping.

A plausible implication is that NSzED’s availability may spur both local clinical translation and broader global efforts toward standardized, multi-ethnic computational psychiatry research (Olateju et al., 2023, Jafari et al., 2023).

In summary, the Nigerian Schizophrenia EEG Dataset (NSzED) fills a crucial gap for both methodological and application-oriented computational neuroscience, providing the foundation for data-driven advances in the diagnosis and understanding of schizophrenia within and beyond African clinical settings.