Enhanced EEG-based Alzheimer's disease detection using synchrosqueezing transform and deep transfer learning.

The most prevalent type of dementia and a progressive neurodegenerative disease, Alzheimer's disease has a major influence on day-to-day functioning due to memory loss, cognitive decline, and behavioral problems. By using synchrosqueezing representations of EEG signals classified by fine-tuned pre-trained convolutional neural networks, this paper presents an EEG-based classification model for Alzheimer's detection. EEG signals are converted into image patterns with time-varying oscillatory elements using the synchrosqueezing technique. The classification performances of the pre-trained deep architectures (SqueezeNet, ResNet, InceptionV3, and MobileNet) using these EEG images are compared. The P3 and T5 channels are the most effective for detecting Alzheimer's disease, according to independent experiments done on EEG signals obtained from 19 scalp electrodes. With classification accuracies of 98.50% and 97.57% for the P3 and T5 channels, respectively, InceptionV3 performs the best. The study also emphasizes that the parietal and temporal lobes' typical disease dynamics are primarily reflected in the electrical activity of the cerebral cortex.