| Abstract [eng] |
Alzheimer’s disease (AD) is the most common form of dementia, associated with gradual decline in cognitive functions, particularly memory and learning. Another common cause of dementia is frontotemporal dementia (FTD), which is characterised by behavioural changes or language impairments. While there are several diagnostic methods for differentiating dementia subtypes, they are expensive, slow and not widely accessible. Therefore, there is a need for more affordable and accessible differential diagnostic tools. The aim of this study was to implement an EEG signal analysis to recognise AD from FTD using statistical and deep learning methods. Resting-state, closed- eyes electroencephalogram (EEG) recordings from individuals diagnosed with AD and FTD were analysed. Spectral analysis was performed using Fast Fourier Transform, and functional connectivity (FC) was calculated using Pearson’s correlation and coherence. The FC matrices were used as input features for a convolutional neural network (CNN) and, for comparison, for a support vector machine (SVM) classifier. The performance of models was evaluated using leave-one-subject-out cross-validation. Feature importance for the CNN classification was obtained using gradient input saliency. The results revealed that theta band power was significantly higher in the AD group compared with the FTD group, while no significant differences were observed in other frequency bands. Furthermore, the AD exhibited a tendency toward higher theta band FC in frontal and posterior temporal regions, while FTD exhibited a tendency toward higher alpha band connectivity in central regions. The CNN-based model achieved higher classification accuracy than the SVM classifier, with accuracies of 85.6% (SD = 25.8) and 57.1% (SD = 35.4). Frontal and occipital region connections were identified as the most important connectivity features for distinguishing between AD and FTD. These results show the potential of EEG features combined with deep learning models as a more affordable and accessible tool for differentiating dementia subtypes. |
