Abstract
Diagnosing Alzheimer's disease (AD) through pathological markers is typically costly and invasive. This study aims to find a noninvasive, cost-effective method using portable electroencephalography (EEG) to detect changes in AD-related biomarkers in cerebrospinal fluid (CSF). A total of 102 patients, both with and without AD-related biomarker changes (amyloid beta and phosphorylated tau), were recorded using a 2-minute resting-state portable EEG. A machine-learning algorithm then analyzed the EEG data to identify these biomarker changes. The results showed that the machine learning model could distinguish patients with AD-related biomarker changes, achieving 68.1% accuracy (AUROC 0.75) for amyloid beta and 71.2% accuracy (AUROC 0.77) for phosphorylated tau, with gamma activities being key features. When excluding cases with idiopathic normal pressure hydrocephalus, accuracy improved to 74.1% (AUROC 0.80) for amyloid beta and 73.1% (AUROC 0.80) for phosphorylated tau. This study suggests that portable EEG combined with machine learning is a promising noninvasive and cost-effective tool for early AD-related pathological marker screening, which could enhance neurophysiological understanding and diagnostic accessibility.