CNN-Based EEG Spectrogram Analysis for Accurate Detection of Alzheimer’s and Dementia

Identifying Alzheimer’s Disease (AD) in its initial stages enables prompt medical intervention and improved patient care strategies. Our research introduces an advanced deep learning framework utilizing Convolutional Neural Networks (CNNs) to analyze electroencephalography (EEG) spectrogram patterns, distinguishing between three clinical categories: Alzheimer’s Disease (AD), Frontotemporal Dementia (FTD), and neurologically healthy controls (CN). Utilizing a dataset of EEG recordings from 88 participants, the CNN model was meticulously developed and trained to capture intricate neural patterns associated with cognitive impairments. Comprehensive data preprocessing, including filtering, Independent Component Analysis (ICA), and spectrogram generation, ensured high signal integrity and feature relevance. Performance analysis revealed the CNN’s superior capabilities with 81.09% accuracy, significantly exceeding Logistic Regression (59.46%) and Random Forest (56.16%) approaches. Key performance indicators—Cohen’s Kappa (0.75), AUC scores (AD: 0.88, FTD: 0.82, CN: 0.90), and MCC (0.74) validated the model’s robust classification abilities. However, limitations such as dataset size and variability in EEG acquisition protocols were acknowledged. These findings demonstrate the efficacy of deep learning techniques in enhancing the accuracy and reliability of AD diagnosis through EEG analysis. The implications of this research suggest that integrating CNN-based models into clinical practice can significantly improve early detection rates of Alzheimer’s Disease, thereby facilitating timely interventions and better patient outcomes.