Comprehensive methodology for sample enrichment in EEG biomarker studies for Alzheimer's risk classification
Verónica Henao Isaza1,2,3, David Aguillon2, Carlos Andrés Tobón-Quintero1
1Grupo Neuropsicología y Conducta (GRUNECO), Facultad de Medicina, Universidad de Antioquia (UdeA), Medellín, Colombia.
Plos One
|March 11, 2026
Summary
This study introduces a novel framework using electroencephalography (EEG) to improve Alzheimer's disease (AD) risk classification. Harmonization and sample enrichment enhance accuracy and generalizability for early AD detection.
Area of Science:
- Neuroscience
- Biomarkers
- Machine Learning
Background:
- Dementia, especially Alzheimer's disease (AD), is a significant global health challenge.
- Electroencephalography (EEG)-based biomarkers offer potential for early AD risk identification.
- Small, heterogeneous samples in EEG studies limit the generalizability of findings.
Purpose of the Study:
- To develop an EEG-based sample enrichment framework to overcome limitations in AD risk classification.
- To improve the generalizability and stability of classification models using multicenter EEG data.
- To address sample size and variability constraints in early AD detection research.
Main Methods:
- Developed an EEG sample enrichment framework integrating signal processing, feature extraction, neuroHarmonize, and Propensity Score Matching (PSM).
- Harmonized EEG data from four cohorts to reduce site variability while retaining covariates like age and sex.
- Extracted features (power, entropy, coherence, etc.) and applied PSM at various ratios to balance control and Alzheimer's risk groups.
Main Results:
- Sample enrichment via PSM significantly improved classification accuracy, with decision tree models achieving 0.91-0.96 accuracy.
- Higher enrichment ratios led to enhanced model stability and generalizability, confirmed by learning curves and confusion matrices.
- Feature selection was guided by model performance and effect sizes (Cohen's d).
Conclusions:
- The proposed framework effectively addresses sample size and variability issues in EEG-based AD risk classification.
- Data harmonization and statistical balancing offer a reproducible strategy for multicenter EEG studies focused on early AD detection.
