For the millions of adults at risk of Alzheimer's disease, the window for meaningful intervention is narrow — and current diagnostic tools either arrive too late or require invasive procedures. A study published in Alzheimer's Research & Therapy now demonstrates that non-invasive brainwave recordings, analyzed through machine learning, can reliably distinguish early Alzheimer's from normal aging, while also tracking biological markers deep within the brain.

Researchers recruited 101 individuals with mild cognitive impairment or mild Alzheimer's disease, alongside 69 age- and education-matched healthy controls. Two EEG-derived feature sets were examined: power spectral density (PSD), which captures the frequency-band composition of neural oscillations, and microstate analysis, which characterizes the brief, stable topographical configurations the brain cycles through at rest. Five machine learning algorithms were trained and compared. A logistic regression model combining both EEG feature types achieved the strongest performance, with a mean area under the ROC curve of 0.859 — suggesting robust discrimination across the early disease spectrum. SHAP analysis identified the most diagnostically influential EEG variables, and correlation work linked these features to cerebrospinal fluid pathological biomarkers, with mediation analysis suggesting EEG signatures partially explain cognitive decline through CSF-reflected amyloid and tau pathology.

This work sits at a productive intersection of two accelerating fields: EEG-based neuroscience and clinical AI. Slowing neural oscillations — particularly reduced alpha power and disrupted theta dynamics — have been associated with AD for decades, but translating these observations into reliable clinical tools has been elusive. Using ensemble feature selection (sequential forward selection) rather than arbitrary frequency-band cutoffs adds methodological rigor. That said, the cohort is modest at 170 participants, models were not externally validated on independent datasets, and MCI does not uniformly progress to Alzheimer's. The link to CSF biomarkers is intriguing but correlational, not mechanistic. Still, the vision of a scalable, low-cost EEG screening tool that mirrors invasive biomarker status — without a lumbar puncture — represents a clinically meaningful direction that warrants larger replication studies.