Slow cortical potential signal classification using concave-convex feature.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Slow cortical potential (SCP) signal classification is vital for disease diagnostics, human-machine interaction, and education.
• Existing multi-feature classification methods lack robustness to changing scenarios.

Purpose of the Study:

• To introduce a novel, flexible concave-convex (C-C) feature for improved SCP signal classification.
• To evaluate the performance and scalability of the proposed C-C feature.

Main Methods:

• Wavelet packet decomposition to extract low-frequency node coefficients from SCP signals.
• Third-order polynomial fitting to estimate the trend of coefficients.
• Feature construction using minimum and maximum second derivative values (|y_min|-y_max).

Main Results:

• The single C-C feature achieved high classification accuracies (92.5% and 84.9%) on benchmark datasets.
• Accuracy improved to 94.5% and 85.9% when combined with the mean voltage feature and a naive Bayesian classifier.
• The proposed method demonstrated superior performance compared to state-of-the-art multi-feature techniques.

Conclusions:

• The concave-convex (C-C) feature is effective for SCP classification.
• The proposed method offers a valuable contribution, balancing traditional and novel feature strengths for SCP analysis.