WMsorting: Wavelet Packets Decomposition and Mutual Information Based Spike Sorting Method.

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Advancements in extracellular multielectrodes enable multi-channel neural recordings.
• Accurate spike sorting is crucial for distinguishing individual neuron activity from dense neural data.
• Existing spike sorting algorithms struggle with noise and overlapping neural signals.

Purpose of the Study:

• To develop a highly accurate, semi-supervised spike sorting algorithm.
• To address challenges posed by noise and overlapping spikes in neural recordings.
• To automate the spike sorting framework with a short training phase.

Main Methods:

• Proposed a novel spike feature extraction method.
• Utilized Wavelet Packets Decomposition (WPD) for feature extraction.
• Employed Mutual Information (MI) for enhanced feature discrimination.
• Implemented a semi-supervised learning approach for spike sorting automation.

Main Results:

• Achieved high clustering accuracy up to 99.76% on public datasets.
• Demonstrated effectiveness in distinguishing individual neuron activities.
• Showcased robust performance against varying noise levels (5-20%).
• Successfully handled data with significant overlapping spikes.

Conclusions:

• The proposed WPD and MI-based spike sorting method is effective and accurate.
• The algorithm offers a promising solution for automated neural signal processing.
• The approach provides good anti-noise performance and high clustering accuracy compared to state-of-the-art methods.