Quantum enhanced EEG classifier towards brain-controlled wheelchair navigation.

Area of Science:

• Neuroscience
• Computer Science
• Quantum Computing

Background:

• Brain-computer interfaces (BCIs) enable assistive technologies but face challenges in accurate electroencephalography (EEG) signal classification due to noise and variability.
• Motor imagery (MI) classification is crucial for controlling devices like brain-controlled wheelchairs.

Purpose of the Study:

• To develop a hybrid Quantum Enhanced CNN-LSTM model (HQeCL) for improved EEG-based MI classification.
• To integrate diverse EEG features (frequency, spatial, non-linear) and quantum-inspired techniques for robust classification.

Main Methods:

• Proposed a hybrid model (HQeCL) combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) with a simulated quantum pooling layer.
• Integrated Power Spectral Density (PSD), Common Spatial Patterns (CSP), and quantum entropy for comprehensive feature extraction.
• Evaluated using leave-one-subject-out (LOSO) cross-validation on an 8-channel MI dataset.

Main Results:

• Achieved high performance metrics: 92.1%±5.9 accuracy, 93.1%±6.2 precision, 91.9%±1.3 recall, 92.5%±1.3 F1-score, and Cohen's κ=0.89±0.02.
• Outperformed existing methods like CSP-LDA, ShallowConvNet, and CNN-LSTM, demonstrating competitive results with QuEEGNet.
• Ablation studies confirmed the benefits of quantum pooling (+0.7% accuracy) and UMAP for feature reduction.
• Demonstrated computational efficiency with 0.12M parameters, 270.2M FLOPs, and 77.6ms inference latency.

Conclusions:

• The HQeCL model offers a quantum-inspired, efficient, and high-performing solution for EEG-based motor imagery decoding.
• The approach shows near real-time feasibility in simulation, advancing the potential for advanced brain-controlled assistive technologies.
• Further research is needed for hardware implementation to fully realize the potential of this BCI advancement.