EEG-Based Alzheimer's Disease Recognition Using Robust-PCA and LSTM Recurrent Neural Network.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalography (EEG) is increasingly used for diagnosing neurodegenerative diseases like Alzheimer's disease (AD).
• Machine learning (ML) offers improved accuracy and data handling for AD recognition compared to manual methods.
• ML methods struggle with incomplete or corrupted data, necessitating robust preprocessing techniques.

Purpose of the Study:

• To develop an automatic classification method for EEG data that remains effective despite artifacts and data corruption.
• To evaluate the performance of Robust Principal Component Analysis (RPCA) in handling corrupted EEG data for AD detection.
• To demonstrate the superiority of RPCA over traditional Principal Component Analysis (PCA) in a machine learning context.

Main Methods:

• Developing a recurrent neural network (RNN) for EEG data analysis.
• Implementing Robust Principal Component Analysis (RPCA) to pre-filter EEG signals corrupted by artifacts and missing data.
• Comparing the performance of an RNN using RPCA-filtered data against one using PCA-filtered data.

Main Results:

• RPCA successfully filtered outlier components from corrupted EEG data.
• The RNN model utilizing RPCA demonstrated increased detection accuracy by approximately 5% compared to the baseline PCA.
• The method showed effectiveness even with up to 20% data erasures.

Conclusions:

• RPCA is a robust technique for preprocessing artifact-affected EEG data, improving ML model performance.
• This approach enhances the reliability of automated AD detection using EEG, particularly in challenging data conditions.
• The study highlights the potential of combining RPCA with RNNs for accurate neurodegenerative disease diagnosis.