Brain wave patterns could reveal early cognitive decline before dementia sets in

An analysis of brain electrical activity classified mild cognitive impairment with high accuracy, pointing toward noninvasive screening options.

Subtle shifts in memory and processing speed often precede dementia by years, yet identifying this transitional stage remains difficult in routine clinical practice. Mild cognitive impairment represents a measurable decline in cognitive function that does not yet prevent independent daily living, though it frequently progresses to Alzheimer disease or other forms of dementia.¹ Pinpointing these subtle cognitive changes early gives clinicians an opportunity to deliver supportive therapies and monitor at-risk patients before extensive neurological damage takes place.

Standard clinical assessments for cognitive decline typically depend on lengthy neuropsychological questionnaires and specialized brain imaging scans.¹ While established cognitive batteries and imaging studies provide valuable structural insights, their cost, duration, and equipment requirements limit their availability for routine, widespread screening. In contrast, electroencephalography records electrical fluctuations along the scalp with high temporal precision, capturing immediate brain activity as millions of neurons communicate through rhythmic electrical pulses.

A study published on August 24, 2026, in the journal Dementia & Neuropsychologia by Hemlata Sandip Ohal and Shamla Mantri evaluates whether deep learning models can recognize mild cognitive impairment directly from brain wave recordings.¹ By analyzing temporal patterns in electrical brain activity, the researchers tested an automated classification system designed to distinguish between individuals with mild cognitive impairment and healthy control subjects.

How do brain wave patterns reflect cognitive decline?

Changes in the electrical rhythms produced by firing neurons reveal early disruptions in brain connectivity that occur as cognitive impairment develops. When groups of neurons communicate across different brain regions, they generate oscillating electrical signals that travel to the scalp. In a healthy brain, these oscillations follow coordinated rhythms that shift predictably depending on mental activity, rest, and sensory processing. As neurodegenerative processes begin to disrupt synaptic connections, the timing and coordination of these electrical pulses alter, creating distinct signal variations over time.

Tracking these subtle timing shifts requires analyzing continuous sequences of electrical activity rather than static snapshots. Neural networks designed for sequential data can learn the temporal relationships across consecutive segments of brain wave signals. By evaluating how voltage patterns fluctuate from one moment to the next, computational models can learn subtle signatures associated with cognitive changes that human observers or static algorithms might miss.

What did the automated classification model achieve?

The 64-node neural network achieved an overall accuracy of 98.14% when classifying segmented brain wave recordings from subjects with mild cognitive impairment against healthy controls.¹ In the study by Ohal and Mantri, the deep learning model evaluated a publicly available dataset of 27 participants, consisting of 11 individuals diagnosed with mild cognitive impairment and 16 cognitively healthy subjects.¹ The model produced a precision of 99.21%, a recall of 98.69%, and an F1-score of 98.95% on the segmented data.¹

The classification performance surpassed traditional machine learning approaches tested on the same data, including K-Nearest Neighbors and Support Vector Machine algorithms.¹ When evaluating individual data segments, the confusion matrix reported by the authors showed that the model correctly identified 4,774 normal segments and 3,257 mild cognitive impairment segments, with few classification errors across the segmented evaluations.¹

To prepare the electrical recordings for the neural network, Ohal and Mantri applied a standard preprocessing pipeline to clean the raw signals.¹ The procedure used band-pass filtering to isolate relevant frequency bands, Independent Component Analysis to identify and remove artifacts such as muscle movements and eye blinks, and segmentation to divide continuous recordings into structured temporal windows.¹ These processed segments were then fed into a 64-node long short-term memory network, a specialized recurrent neural architecture designed to retain temporal dependencies over time.¹

What are the limits of testing on small subject cohorts?

The findings represent an initial demonstration of algorithm performance on a small experimental group rather than a validated clinical diagnostic test. Because the evaluation relied on a dataset of 27 individuals, the high reported accuracy reflects performance on a limited sample that may not capture the biological and clinical diversity of broader patient populations. The high segment counts in the evaluation were generated by dividing recordings from these 27 participants into thousands of shorter temporal windows, meaning the statistical metrics evaluate segment-level classification rather than independent patient-level assessments across diverse clinical cohorts.

The study also evaluated a binary classification task between distinct healthy individuals and patients already diagnosed with mild cognitive impairment. In clinical settings, patients often present with overlapping symptoms from medication side effects, metabolic conditions, mood disorders, or other neurological changes that complicate diagnosis. Testing the model across larger, independent patient populations from multiple recording sites will be necessary to determine how well the classification generalizes beyond the specific public dataset analyzed.

What steps follow for electroencephalogram analysis?

Future development will require evaluating deep learning models on prospective patient groups across diverse healthcare settings to measure real-world screening performance. Automated brain wave analysis could potentially support noninvasive, scalable cognitive assessments if models demonstrate reliable performance across varied recording hardware and diverse patient demographics. Refining computational pipelines for scalp-recorded brain activity offers a path toward exploring low-cost diagnostic support tools for early cognitive monitoring.

This piece was prepared from the paper published in Dementia & Neuropsychologia and public records; the authors have not been interviewed.

References

This article is based on 1 source, listed in the order they are cited.

1. 1 HS Hemlata Sandip Ohal, Shamla Mantri announcement · 26 Aug 2026 A deep learning approach to mild cognitive impairment detection from electroencephalogram signals See the source

Article history

1. Update 27 Aug 2026, 04:15

   This article was updated to reflect revised text and images.

2. Published 27 Aug 2026, 04:16
   Assembled by the Primary desk from 1 source · 11 cited sentences