Quantum neural network-based EEG filtering for a brain-computer interface.

Area of Science:

• Quantum computing and artificial intelligence
• Signal processing and machine learning
• Neuroscience and brain-computer interfaces

Background:

• Traditional signal processing methods struggle with nonstationary stochastic signals embedded in noise.
• Existing neural networks may not fully capture the complex dynamics of biological signals.
• The Schrodinger wave equation offers a novel framework for information processing.

Purpose of the Study:

• To introduce a novel recurrent quantum neural network (RQNN) architecture.
• To develop an unsupervised learning algorithm for RQNNs to capture signal behavior.
• To evaluate RQNN performance in filtering noisy signals and improving brain-computer interface (BCI) applications.

Main Methods:

• Proposed a recurrent quantum neural network (RQNN) inspired by quantum mechanics and the Schrodinger wave equation.
• Utilized an unsupervised learning algorithm for RQNN to characterize nonstationary signals as time-varying wave packets.
• Applied RQNN for filtering electroencephalogram (EEG) signals in a motor imagery-based BCI, using particle swarm optimization and cross-validation for parameter selection.

Main Results:

• RQNN accurately filtered simple signals (dc, staircase dc, sinusoidal) embedded in high noise.
• Subject-specific RQNN filtering significantly improved BCI performance compared to raw or Savitzky-Golay filtered EEG.
• Demonstrated effective estimation of signals with unknown noise characteristics.

Conclusions:

• The recurrent quantum neural network (RQNN) provides a robust method for filtering complex signals in noisy environments.
• RQNN-based filtering enhances signal separability, leading to improved performance in brain-computer interfaces.
• Quantum-inspired neural networks offer a promising direction for advanced signal processing and BCI applications.