Iterative multiblock framework for high frequency EEG based neurological disorder detection.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Early and accurate diagnosis of neurological diseases like Alzheimer's and Parkinson's is crucial.
• High-frequency electroencephalogram (EEG) signals offer potential but face challenges due to noise and non-stationarity.
• Existing diagnostic methods struggle with signal processing, feature selection, fusion, and clinical explainability.

Purpose of the Study:

• To propose a holistic framework for early clinical detection of neurological disorders using enhanced high-frequency EEG signals.
• To overcome limitations in current diagnostic techniques for neurological conditions.

Main Methods:

• A multi-block pipeline combining Hilbert-Huang Transform (HHT) with empirical mode decomposition for adaptive preprocessing and noise reduction.
• Wavelet Packets Transform (WPT) with Shannon entropy for feature selection, preserving temporal and frequency domain information.
• Canonical Correlation Analysis (CCA) for integrating EEG features with clinical metadata, and a Multi-Scale Convolutional Recurrent Neural Network (MS-CRNN) with attention for spatiotemporal analysis.

Main Results:

• The proposed framework achieved 94% accuracy, 92% sensitivity, and 93% specificity in early issue identification.
• Visualization techniques (Grad-CAM, Integrated Gradients) were used for feature attribution and explainability.
• The method effectively extracts clinically relevant features from noisy high-frequency EEG data.

Conclusions:

• The developed framework significantly enhances the accuracy and sensitivity of early neurological disease diagnosis.
• The approach provides a new benchmark for clinical interpretation and diagnosis, supporting early intervention policies.
• The integration of signal processing, feature engineering, and deep learning offers a robust solution for complex neurological disorder detection.