How Is an EEG Used for a Dementia Diagnosis?

An electroencephalogram, or EEG, contributes to dementia diagnosis by detecting shifts in brain electrical activity that track with cognitive decline. It does not give a standalone “yes or no” answer the way a blood test confirms an infection, but it picks up abnormal patterns in brain waves that help clinicians distinguish between types of dementia, rule out other conditions that mimic cognitive decline, and, increasingly, flag early changes before symptoms become obvious. In some clinical comparisons, quantitative EEG has matched or slightly outperformed MRI in identifying Alzheimer’s disease, and it costs a fraction of the price.

What EEG Actually Detects in Dementia

A healthy brain at rest produces a mix of electrical oscillations at different speeds. The ones most relevant to dementia fall into broad frequency bands: slow waves (delta and theta) and faster waves (alpha and beta). In a person with Alzheimer’s disease, the overall pattern shifts toward slower activity. Relative delta and theta power go up, while alpha and beta power go down, compared to both healthy older adults and people with other conditions like type 2 diabetes.¹ This slowing is one of the most consistently replicated EEG findings in dementia research.

The dominant resting brain rhythm, which in healthy older adults typically hums along at about 9 Hz, drops to around 7 to 8 Hz in Alzheimer’s and can dip even lower in dementia with Lewy bodies (DLB), where estimates place it around 6 to 7 Hz.² That difference matters practically: a clinician relying on standard frequency cutoffs might miss the slowing entirely if the patient’s rhythm has drifted just below the typical alpha range. Newer quantitative approaches adjust for this by finding each person’s individual peak frequency rather than relying on fixed boundaries.

Telling Dementia Types Apart

One of EEG’s most clinically useful roles is helping distinguish between the major forms of dementia, which can look similar on a bedside cognitive exam but produce recognizably different electrical signatures.

Alzheimer’s disease shows a characteristic pattern of increased slow-wave activity (delta and theta) with a loss of posterior alpha power. Frontotemporal dementia (FTD) looks different: alpha activity tends to be preserved, but there are reductions in frontal beta activity instead. Dementia with Lewy bodies produces an exaggerated version of the slowing seen in Alzheimer’s, plus a unique reduction in gamma-range activity that neither Alzheimer’s nor FTD typically shows.³

The distinction between Alzheimer’s and DLB is especially important because the two conditions call for different treatment strategies, and DLB patients can have dangerous reactions to certain antipsychotic medications. EEG helps here: DLB patients tend to have a markedly lower peak frequency (around 7 Hz compared to about 8 Hz in Alzheimer’s), more slow-wave activity overall, and greater disruption of brain network connectivity.⁴ Another distinguishing feature is that DLB patients show much wider variability in their dominant frequency over time, fluctuating by nearly twice as much as Alzheimer’s patients during a recording session.⁵ That fluctuation tracks with the clinical reality of DLB, where cognition and alertness shift unpredictably from hour to hour.

Vascular dementia shares some features with Alzheimer’s, including increased delta power and a slowing of the alpha rhythm, but with a key difference: alpha power actually tends to stay higher in vascular dementia than in Alzheimer’s, particularly over the back of the head.⁶ Both conditions show a decrease in beta power and wider variability in alpha frequency compared to healthy controls, but the relative preservation of alpha in vascular dementia gives clinicians a distinguishing marker.

FTD versus Alzheimer’s is another pairing where EEG adds value. Both spectral analysis and connectivity mapping can differentiate the two at a group level, though distinguishing individual patients remains harder. Combining power measurements with source localization techniques improves sensitivity.⁷

Connectivity and Network Breakdown

Beyond just looking at how fast or slow brain waves are, modern EEG analysis examines how well different brain regions communicate with each other. In Alzheimer’s, this functional connectivity declines progressively with disease severity, particularly in the alpha band.⁸ The synchronized communication between frontal and parietal areas, which supports attention and working memory, weakens as the disease advances. The loss is measurable between specific brain regions, including reduced synchronization between prefrontal and parietal areas in the theta band.⁹

This connectivity dimension is useful because it captures something different from the raw power measurements. Two patients could have similar amounts of slowing but very different patterns of network disruption, which may reflect different disease stages or different underlying pathologies. It is also one of the features that helps predict who will progress from mild cognitive impairment to full dementia, which is covered further below.

Microstates and Brief Brain Patterns

A newer layer of EEG analysis looks at “microstates,” which are brief, stable patterns of electrical activity across the scalp that last only fractions of a second before switching to another pattern. Researchers have identified a handful of canonical microstate classes, labeled A through D, and Alzheimer’s patients show a characteristic disorganization in how these microstates behave.¹⁰

The specifics get interesting. In Alzheimer’s, certain microstates linger longer than they should, and the normal sequence in which the brain transitions between states becomes scrambled. One study found that the transition probability between two specific microstate classes correlated negatively with cognitive test scores, meaning worse cognition tracked with more frequent abnormal transitions.¹¹ Another study linked specific microstate changes to the degree of amyloid-beta deposition in the brain, suggesting that these fleeting electrical patterns might serve as a window into underlying Alzheimer’s pathology.¹² Microstate analysis is still largely a research tool, but it represents a direction where EEG could eventually contribute more fine-grained diagnostic information than simple frequency analysis alone.

Predicting Who Will Progress from Mild Cognitive Impairment

Perhaps the highest-stakes clinical question is not whether someone already has dementia but whether a person with mild cognitive impairment (MCI) will convert to dementia in the coming years. Several EEG features have shown promise here. One study using source analysis and coherence measurements found that stronger temporal delta activity and higher midline gamma coherence at baseline were associated with dramatically higher annual conversion rates, reaching up to 40 to 60 percent compared to around 10 percent in those without these markers.¹³

Connectivity measures also contribute. Regional differences in beta-band connectivity between central and lateral brain areas have shown some ability to classify which MCI patients will eventually develop dementia, with specificity reaching the mid-80s in percentage terms.¹⁴ Machine learning approaches applied to high-density EEG recordings have pushed individual-level classification accuracy even higher in small studies, though those results need replication in larger groups before they change clinical practice.¹⁵

The appeal here is obvious: EEG is cheap, widely available, and repeatable. If an annual EEG could flag the patients most likely to progress, clinicians could target those individuals for closer monitoring, earlier interventions, or enrollment in clinical trials for emerging therapies.

The P300 and Other Event-Related Potentials

Standard clinical EEG records brain activity at rest. But a related technique involves measuring the brain’s response to specific stimuli, called event-related potentials (ERPs). The most studied of these in dementia is the P300, a positive voltage deflection that appears about 300 milliseconds after a person hears or sees an unexpected stimulus. In Alzheimer’s and MCI, the P300 arrives later than it should. This delay has been proposed as a quantitative, unbiased marker of cognitive change that can track progression from mild to moderate impairment and potentially detect treatment effects.¹⁶

Gamma-frequency responses to sensory stimulation, particularly flashing light, also differ between groups. Compared to healthy older adults, people with MCI and Alzheimer’s show impaired sensory-evoked gamma responses, suggesting that the brain’s ability to synchronize high-frequency activity in response to stimulation deteriorates as cognitive decline advances.¹⁷ This gamma research has a therapeutic angle as well: there is active investigation into whether external gamma-frequency stimulation might slow Alzheimer’s progression, though that work is still in early stages.

Sleep EEG Markers

Some of the earliest EEG abnormalities in Alzheimer’s may appear not during waking hours but during sleep. Sleep spindles, brief bursts of rhythmic brain activity that occur during lighter stages of sleep and are thought to play a role in memory consolidation, decline in both density and location-specificity in Alzheimer’s patients. Specifically, fast sleep spindle density over the parietal region decreases in both Alzheimer’s and MCI, and the degree of that decrease correlates with cognitive test performance.¹⁸ The fact that spindle changes are already present in MCI suggests they could serve as an early biomarker, detectable before full-blown dementia sets in.

How EEG Compares to MRI

The natural question most people ask is: why bother with EEG when you can get a brain scan? In a community-based study comparing the two, visual EEG assessment reached about 81 percent total accuracy in diagnosing Alzheimer’s, matching or slightly exceeding MRI’s 72 percent accuracy based on medial temporal lobe atrophy measurements. Quantitative EEG pushed that figure to 81 to 84 percent. Interestingly, the patients identified by each method did not completely overlap, suggesting that EEG and MRI capture different aspects of the disease and work best as complements rather than competitors.¹⁹

EEG also has practical advantages. It is considerably less expensive than MRI, and it measures brain function in real time rather than structure at a single moment. For patients who cannot tolerate an MRI scanner due to claustrophobia, metallic implants, or frailty, EEG offers a viable alternative. A cost survey noted that EEG is more than six times cheaper than MRI for outpatient neurological examination.²⁰

Machine Learning and Automated Classification

The human eye can catch gross EEG abnormalities, but the subtle spectral and connectivity shifts in early dementia are often invisible on visual inspection. This is where computerized quantitative EEG and machine learning enter the picture. In a proof-of-concept study using support vector machine classifiers trained on resting EEG features, classification accuracy reached about 98 percent for distinguishing healthy older adults from Alzheimer’s patients, nearly 100 percent for healthy versus DLB, and about 98 percent for Alzheimer’s versus DLB. Even a three-way classifier separating all three groups achieved roughly 95 percent accuracy.²¹

Those numbers are striking but need context. Proof-of-concept studies typically use small, well-characterized samples under controlled conditions. Real-world clinical settings introduce noise, comorbidities, medication effects, and the general mess of everyday medicine. Artifact contamination from muscle tension, eye blinks, and electrical interference can degrade EEG signals, and automated artifact removal is a research area in its own right. Studies have tested several cleaning algorithms and found that wavelet-enhanced independent component analysis outperformed other methods for preserving diagnostically useful signal while removing noise.²² Getting artifact rejection right matters enormously, because if your algorithm is classifying noise rather than brain activity, the real-world accuracy will be far lower than the lab accuracy.

Wearable EEG and the Push Toward Screening

Traditional clinical EEG requires a technician to apply 20 or more electrodes with conductive gel, which takes time and needs to happen in a clinic. Wearable EEG devices, using dry electrodes and wireless transmission, are being developed as screening tools that could eventually be used in a primary care office or even at home. A systematic review of wearable EEG for detecting MCI found that classification accuracy across studies ranged from 46 to 95 percent, with a median in the mid-70s to mid-80s. Systems using four to eight channels performed significantly better than those with only one to three, but going beyond nine channels did not meaningfully improve results.²³

The wide accuracy range reflects how much the result depends on electrode placement, the cognitive task the patient performs during recording, signal processing choices, and the classifier used. Frontal and parietal electrode positions emerged as the most informative, and combining EEG with other data sources (like cognitive task performance) boosted results. The technology is still maturing, but the trajectory is clear: cheaper, simpler EEG devices that can bring screening to people who would never be referred to a specialist neurophysiology lab.

Monitoring Treatment Response

EEG is not only useful at the point of diagnosis. It can also track how a patient’s brain responds to medication. In patients taking donepezil, one of the standard cholinesterase inhibitor drugs for Alzheimer’s, EEG recordings after six months showed reduced delta activity and increased alpha and beta activity, essentially a partial reversal of the disease-related slowing pattern.²⁴ Not every patient responds the same way, and EEG may help distinguish responders from non-responders. One study found that patients who responded clinically to donepezil showed a specific restoration of temporal and occipital alpha rhythms that was absent in non-responders, suggesting the drug reactivated dormant cortical circuits in those who benefited.²⁵

This has practical implications as families and doctors try to decide whether a medication is actually helping. Cognitive tests can fluctuate day to day depending on sleep, mood, and effort, but EEG changes provide a more objective physiological measure that could supplement subjective assessments.

Ruling Out Delirium and Creutzfeldt-Jakob Disease

Two situations where EEG plays an especially clear-cut diagnostic role deserve mention. The first is delirium, a sudden, fluctuating confusion that can look like dementia but has entirely different causes (infections, medication reactions, metabolic problems). A normal resting EEG makes the presence of delirium very unlikely, which helps clinicians avoid misdiagnosing a treatable acute condition as irreversible cognitive decline.²⁶

The second is Creutzfeldt-Jakob disease (CJD), a rare but rapidly fatal prion disease that can initially mimic other dementias. CJD produces a distinctive EEG pattern of periodic sharp-wave complexes, repeating spike-like discharges that recur at roughly regular intervals. In a large study, these complexes appeared in 64 percent of confirmed CJD cases and were falsely positive in only 9 percent of other dementias (mostly Alzheimer’s), yielding a specificity of 91 percent and a positive predictive value of 95 percent.²⁷ When they are present, these periodic complexes also identify a particular phenotype of CJD characterized by faster disease progression and shorter survival time.²⁸ In a clinical setting where dementia is advancing unusually fast, EEG is often one of the first tests ordered precisely because of its ability to flag CJD.

Where EEG Fits in the Bigger Diagnostic Picture

EEG does not replace a careful clinical history, cognitive testing, blood work, or brain imaging. What it adds is a functional readout of how the brain’s electrical networks are actually performing, which complements the structural information from an MRI or CT scan. In practice, EEG is most often ordered when there is diagnostic uncertainty: when symptoms could point to more than one type of dementia, when delirium needs to be excluded, when CJD is on the differential, or when a clinician wants an objective measure to track over time. Its low cost and repeatability make it particularly suited for serial monitoring, something that is impractical with repeated MRIs or PET scans.

The research trajectory points toward EEG becoming more automated, more portable, and more embedded in routine care. Machine learning classifiers trained on EEG data are approaching the accuracy of imaging-based methods in controlled settings, wearable devices are shrinking the technology to a few well-placed electrodes, and the underlying neuroscience keeps revealing new EEG features that track with disease biology. For now, EEG is a supporting player in the dementia diagnostic toolkit. But it is one of the few tools that can measure brain function directly, inexpensively, and repeatedly, which gives it a role that brain scans and blood tests cannot fully replicate.