Spike detection and clustering with unsupervised wavelet optimization in extracellular neural recordings

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Accurate detection of action potentials in noisy neural recordings is crucial for brain-computer interfaces.
• Existing methods often struggle with unknown waveforms and high noise levels.

Purpose of the Study:

• To develop a robust and accurate method for detecting action potentials in noisy extracellular neural recordings.
• To improve spike sorting performance by optimizing wavelet selection.

Main Methods:

• Introduced a novel wavelet-based manifestation variable combining wavelet shrinkage denoising and multiscale edge detection.
• Implemented unsupervised optimization for best basis selection to eliminate mother wavelet dependence.
• Defined an unsupervised criterion based on correlation similarity for updating wavelet selection during clustering.

Main Results:

• The proposed method significantly outperformed previous methods in detecting action potentials across various simulated and experimental datasets.
• Updating wavelet selection during clustering improved classification performance compared to using a fixed wavelet.

Conclusions:

• The new wavelet-based method offers a robust solution for action potential detection in noisy neural signals.
• Unsupervised optimization of wavelet selection enhances both detection accuracy and spike sorting performance for brain-computer interfaces.