PET‐validated EEG‐machine learning algorithm predicts brain amyloid pathology in pre‐dementia Alzheimer’s disease

Background Currently positron emission tomography (PET) is used as the initial or sole biomarker of β-amyloid (Aβ) brain pathology, which may inhibit Alzheimer’s disease (AD) drug development and clinical use due to cost, access, tolerability. Machine learning (ML)-based EEG biomarkers address these challenges. Previous studies have confirmed their ability accurately discriminate between normal, mild cognitive impairment (MCI) dementia. However, few quantitative (QEEG) been validated with Aβ PET. We developed a QEEG-ML algorithm predict pathology among subjective decline (SCD) MCI patients, it using Method (19-channel, eye-closed, resting-state) PET data were collected from 311 subjects SCD (Aβ+: N = 36, Aβ-: 160) 54, 61). randomly excluded for 76 (SCD, Aβ+: 7, 28; MCI, 6, 7) in subsequent verification. QEEG absolute power, relative power ratio, connectivity channels (iCoherence) comprised input features, most relevant predictive features identified several methods (Random Forest Importance (GBM, XGB), ElasticNet, Whitney-Mann). then trained four ML algorithms (SVM, Logistic, KNN, Naive Bayes, Random Forest(GBM/XGB)) each feature set, yielding 24 models (4 sets * 6 algorithms). The validation into model compare performance. Result best-performing (random forest importance SVM) showed 82.9% accuracy, 90.9% sensitivity, 76.7% specificity, 75% positive value discriminating Aβ+ Aβ-, regardless MCI/SCD. In alone, 82.1% 90.0% 78.9% 81.8% value. showed, 81.1% 92.3% 66.7% Conclusion These findings suggest that our novel ML-based can presence plaque. Additional benefits such include reduced expense, wide availability, high-throughput screening response monitoring. Future will assess utility primary AD screens, adjunctive PET, support rationales treatment choice AD.