Hybrid Quantum-Classical Model That Combines Spatial-Temporal EEG and Digitized Counterdiabatic Quantum Features for...

Area of Science:

• Neuroscience
• Quantum Computing
• Machine Learning

Background:

• Motor imagery (MI) classification using electroencephalogram (EEG) is crucial for brain-computer interfaces.
• Current deep learning methods face challenges in accurately interpreting complex EEG patterns.

Purpose of the Study:

• To develop and evaluate a quantum-feature enhanced algorithm for improved MI classification accuracy.
• To integrate quantum computing principles into EEG signal processing for enhanced feature extraction.

Main Methods:

• A spatial-temporal deep learning architecture was employed to process 58-lead EEG waveforms.
• Digitized counterdiabatic quantum feature extraction was performed on a reduced 24-dimension feature set.
• A hybrid classifier combined deep learning predictions with quantum features, weighting quantum features during model uncertainty.

Main Results:

• The hybrid model achieved 88.8% classification accuracy and 0.962 AUROC on an external evaluation participant.
• Overall classification accuracy across 51 participants reached 89.8% with a 0.970 AUROC.
• Model performance demonstrated a positive correlation with participant age (Spearman's rho = 0.35, p=0.012).

Conclusions:

• The integration of quantum features with deep learning significantly enhances the robustness of MI EEG classification.
• This novel approach establishes a foundation for exploring quantum computing's potential in healthcare applications.
• Further research is warranted to fully leverage quantum-enhanced algorithms in clinical settings.