 |
Electroencephalography (EEG) has been exploited since a long time for Alzheimer's disease (AD) diagnosis. Several studies in the literature investigated functional connectivity to distinguish between AD patients and Healthy controls. In this work, we investigate the impact of analyzing EEG signals with different epoch durations on classification performance, when discriminating AD patients from Subjective Cognitive Impairment (SCI) subjects, using Phase-Lag Index (PLI) to quantify functional connectivity. We find that the PLI measurement is more reliable to distinguish between SCI and AD epochs, when it is estimated on large epochs. Then, going towards the classification of AD and SCI patients, we average the classifier output scores of epochs, for each epoch duration. Results show that fusing the output scores of epochs allows achieving better classification performance, compared to the obtained results on separate epochs. The best classification performance of AD and SCI patients is obtained with epochs of 4 seconds (AUC=0.825, Accuracy=82%). Finally, we propose a new framework based on the fusion of classification results at different epoch durations. Experiments show that this proposal leads to an improvement of classification performance, reaching an AUC of 0.93 and an Accuracy value of 90%, with a good balance between Specificity and Sensitivity.
