An improved catch fish optimization based deep learning model for Parkinson disease classification using EEG signal.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Parkinson's Disease (PD) diagnosis from Electroencephalogram (EEG) signals is challenging due to signal complexity and non-stationarity.
• Manual evaluation of EEG signals for PD is difficult and subjective.
• Objective and automated methods are needed for reliable PD identification.

Purpose of the Study:

• To propose an Improved Catch Fish Optimization (ICFO)-based encoder-decoder deep learning framework for accurate Parkinson's Disease classification using EEG signals.
• To enhance the performance of PD detection by optimizing deep learning hyperparameters.
• To provide a reliable automated tool for EEG-based PD diagnosis.

Main Methods:

• EEG signal preprocessing using a bandpass filter.
• Decomposition of EEG signals into sub-bands using Tunable-Q Wavelet Transform (TQWT).
• Classification using a Multi-Head Attention Bidirectional Long Short-Term Memory (MA-BiLSTM) network optimized by ICFO for hyperparameter tuning.

Main Results:

• The MA-BiLSTM-ICFO model achieved high performance on two public EEG datasets (San Diego and Iowa).
• Achieved accuracy up to 99.7%, recall up to 99.9%, precision up to 99.3%, and F-score up to 99.2%.
• Ablation and comparative studies confirmed the superiority of the proposed framework over existing methods.

Conclusions:

• The proposed ICFO-based encoder-decoder deep learning framework offers a reliable automated approach for Parkinson's Disease classification from EEG signals.
• The integration of TQWT, multi-head attention, BiLSTM, and ICFO significantly improves classification performance.
• This framework has the potential to support clinical diagnosis of PD.