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

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease is a progressive neurodegenerative disorder impacting memory, cognition, and behavior.
• Early and accurate detection is crucial for managing the disease's impact on daily functioning.

Purpose of the Study:

• To develop and evaluate an electroencephalogram (EEG)-based classification model for Alzheimer's disease detection.
• To compare the performance of different pre-trained convolutional neural networks (CNNs) for classifying EEG signals.

Main Methods:

• EEG signals were transformed into image patterns using the synchrosqueezing technique.
• Fine-tuned pre-trained CNNs (SqueezeNet, ResNet, InceptionV3, MobileNet) were used to classify these EEG images.
• Classification performance was evaluated using signals from 19 scalp electrodes, focusing on P3 and T5 channels.

Main Results:

• The InceptionV3 model demonstrated the highest classification accuracy, achieving 98.50% for the P3 channel and 97.57% for the T5 channel.
• The P3 and T5 channels were identified as the most effective for Alzheimer's detection.
• The study confirmed that electrical activity in the parietal and temporal lobes reflects typical disease dynamics.

Conclusions:

• The proposed EEG-based classification model using synchrosqueezing and CNNs shows significant potential for accurate Alzheimer's disease detection.
• The findings highlight the efficacy of specific EEG channels (P3, T5) and the InceptionV3 architecture for this diagnostic task.
• This approach offers a non-invasive method to identify Alzheimer's, reflecting underlying cortical electrical activity patterns.