Clinically Calibrated Machine Learning Benchmarks for Large-Scale Multi-Disorder EEG Classification

TL;DR

This study tackles automated EEG-based screening across 11 clinically meaningful neurological disorders using real-world hospital data. It introduces a clinically calibrated pipeline that processes a 16-channel longitudinal bipolar montage into 3,108 multi-domain features, with disorder-specific classifiers and thresholds tuned to maximize recall. Through XGBoost and MLP implementations, the approach achieves recall above 0.75 for all disorders and above 0.80 for most, providing interpretable feature biomarkers aligned with established EEG signatures. The work demonstrates that sensitivity-oriented, threshold-adjusted multi-disorder EEG analysis can serve as scalable screening and triage support in busy clinical neurophysiology settings, advancing beyond single-disorder seizure detection.

Abstract

Clinical electroencephalography is routinely used to evaluate patients with diverse and often overlapping neurological conditions, yet interpretation remains manual, time-intensive, and variable across experts. While automated EEG analysis has been widely studied, most existing methods target isolated diagnostic problems, particularly seizure detection, and provide limited support for multi-disorder clinical screening. This study examines automated EEG-based classification across eleven clinically relevant neurological disorder categories, encompassing acute time-critical conditions, chronic neurocognitive and developmental disorders, and disorders with indirect or weak electrophysiological signatures. EEG recordings are processed using a standard longitudinal bipolar montage and represented through a multi-domain feature set capturing temporal statistics, spectral structure, signal complexity, and inter-channel relationships. Disorder-aware machine learning models are trained under severe class imbalance, with decision thresholds explicitly calibrated to prioritize diagnostic sensitivity. Evaluation on a large, heterogeneous clinical EEG dataset demonstrates that sensitivity-oriented modeling achieves recall exceeding 80% for the majority of disorder categories, with several low-prevalence conditions showing absolute recall gains of 15-30% after threshold calibration compared to default operating points. Feature importance analysis reveals physiologically plausible patterns consistent with established clinical EEG markers. These results establish realistic performance baselines for multi-disorder EEG classification and provide quantitative evidence that sensitivity-prioritized automated analysis can support scalable EEG screening and triage in real-world clinical settings.

Figures (15)

• Figure 1: Proposed end-to-end framework for EEG-based multi-disorder classification.
• Figure 2: Illustration of the ACNS longitudinal bipolar montage (“double-banana” configuration). Electrode pairs are grouped into four principal anatomical chains spanning both hemispheres: blue line- Left Lateral (LL) Chain (Fp1–F7, F7–T3, T3–T5, T5–O1), yellow line- Left Parasagittal (LP) Chain (Fp1–F3, F3–C3, C3–P3, P3–O1), red line-Right Parasagittal (RP) Chain (Fp2–F4, F4–C4, C4–P4, P4–O2), and green line-Right Lateral (RL) Chain (Fp2–F8, F8–T4, T4–T6, T6–O2).
• Figure 3: PCA visualizations for seizure-related and peripheral nervous system disorders.
• Figure 4: Class imbalance across target disorders.
• Figure 5: Confusion matrix for Seizure Disorders using XGBoost. High sensitivity (91.2%) is achieved while preserving balanced overall accuracy.