LCADNet: a novel light CNN architecture for EEG-based Alzheimer disease detection

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Alzheimer's disease (AD) diagnosis is challenging due to its progressive nature and current methods' limitations.
• Existing automated methods using electroencephalogram (EEG) signals often lack accuracy and reliability.
• There is a need for efficient and precise automated AD detection techniques.

Purpose of the Study:

• To develop and evaluate a lightweight convolutional neural network (LCADNet) for accurate and rapid Alzheimer's disease detection.
• To compare the performance of LCADNet against pre-trained models using transfer learning for AD detection.
• To assess the generalization capability of LCADNet across multiple datasets.

Main Methods:

• A novel lightweight convolutional neural network (LCADNet) architecture was designed, incorporating convolutional, fully connected, and max-pooling layers.
• EEG data from publicly available datasets were used to train and test the LCADNet model.
• The LCADNet's efficiency and classification performance were compared with four pre-trained models using transfer learning.
• Generalization was evaluated by cross-testing the model on two additional AD detection datasets.

Main Results:

• LCADNet demonstrated the lowest computational complexity (floating point operations and inference time).
• The model achieved the highest classification performance across six standard metrics.
• Cross-testing confirmed LCADNet's strong generalization capabilities on unseen datasets.
• LCADNet achieved an outstanding accuracy of 98.50% in EEG-based AD detection.

Conclusions:

• LCADNet provides a highly accurate and computationally efficient method for Alzheimer's disease detection using EEG signals.
• The developed model outperforms existing EEG-based AD detection approaches.
• LCADNet shows potential as a valuable tool to assist neurologists in clinical diagnosis.