Electroencephalogram-based time-frequency analysis for Alzheimer's disease detection using machine learning

Alzheimer's disease (AD) is the most common form of dementia. The lack of effective prevention or cure makes AD a significant concern, as it is a progressive disease with symptoms that worsen over time.

The aim of this study is to develop an algorithm capable of differentiating between patients with early-stage AD (mild cognitive impairment [MCI]), moderate AD, and healthy controls (C) using electroencephalogram (EEG) signals.

A publicly available EEG database was utilized, with seven EEG recordings selected from each study group (MCI, AD, and C) to ensure a balanced dataset. For each 1-s segment of EEG data, 43 time-frequency features were computed. These features were then compressed over time using 10 statistical measures. Subsequently, 15 classifiers were employed to distinguish between paired groups using a 7-fold cross-validation.

The strategy yielded better results than state-of-the-art methods, achieving a 100% accuracy in both C versus MCI and C versus AD binary classifications. This improvement translated to a 2% increase in accuracy for C versus MCI and a 4% increase for C versus AD, despite a 1.2% decrease in performance for AD versus MCI. In addition, the proposed method outperformed prior work on the same database by 4.8% for the AD versus MCI comparison.

The present study highlights the potential of EEG as a promising tool for early AD diagnosis. Nevertheless, a more extensive database should be used to enhance the generalizability of the results in future work.