Arithmetic Optimization with RetinaNet Model for Motor Imagery Classification on Brain Computer Interface.

Area of Science:

• Neuroscience and Biomedical Engineering
• Artificial Intelligence and Machine Learning

Background:

• Brain-computer interfaces (BCIs) are crucial for individuals with motor disabilities, enabling communication and control via brain signals like electroencephalogram (EEG).
• Existing methods for EEG feature learning require further exploration to enhance motor imagery (MI) classification outcomes using deep learning (DL).

Purpose of the Study:

• To design and evaluate a novel deep learning model, AORNDL-MIC, for improved MI classification in BCIs.
• To leverage advanced signal processing and DL techniques for novel EEG feature representation and extraction.

Main Methods:

• EEG signal preprocessing using Multiscale Principal Component Analysis (MSPCA) for denoising.
• Transformation of 1D EEG signals to 2D time-frequency representations using Continuous Wavelet Transform (CWT).
• Feature extraction via RetinaNet deep learning model, followed by classification using the ID3 classifier.
• Hyperparameter tuning of RetinaNet using the Arithmetical Optimization Algorithm (AOA) to enhance classification efficiency.

Main Results:

• The proposed AORNDL-MIC technique successfully transforms 1D EEG signals into a 2D representation suitable for DL models.
• RetinaNet effectively extracts feature vectors from the processed EEG signals.
• AOA-based hyperparameter optimization significantly improved the classification performance of the AORNDL-MIC technique.
• Experimental results demonstrated superior performance of AORNDL-MIC compared to existing state-of-the-art methods on benchmark datasets.

Conclusions:

• The AORNDL-MIC technique offers a promising approach for enhancing MI classification accuracy in BCIs.
• The integration of MSPCA, CWT, RetinaNet, and AOA provides an effective framework for advanced EEG signal analysis.
• This study contributes to the development of more robust and efficient BCIs for individuals with movement disabilities.