Transparent EEG Analysis: Leveraging Autoencoders, Bi-LSTMs, and SHAP for Improved Neurodegenerative Diseases...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Electroencephalogram (EEG) is a non-invasive tool for brain monitoring.
• Neurodegenerative diseases like Alzheimer's disease (AD) and frontotemporal dementia (FTD) present diagnostic challenges.
• High dimensionality and noise in EEG signals complicate analysis.

Purpose of the Study:

• To develop a deep learning pipeline for classifying EEG signals in AD and FTD.
• To leverage autoencoders for feature extraction and Bi-LSTM networks for temporal pattern analysis.
• To enhance model interpretability using SHapley Additive exPlanations (SHAP).

Main Methods:

• A novel classification pipeline combining autoencoders and bidirectional long short-term memory (Bi-LSTM) networks.
• Dimensionality reduction using autoencoders to preserve key EEG features.
• Temporal pattern analysis using Bi-LSTM on resting-state EEG data.

Main Results:

• Achieved 98% accuracy in classifying EEG signals for AD and FTD.
• Successfully identified subtle temporal patterns indicative of neurodegenerative diseases.
• SHAP analysis provided insights into feature contributions for model predictions.

Conclusions:

• The proposed deep learning approach offers a transparent and accurate method for EEG-based AD and FTD classification.
• This technique shows promise for early and cost-effective diagnosis of neurodegenerative diseases.
• Addressing EEG challenges like noise and variability is crucial for clinical translation.