Neuronal Waveform Classification in Multielectrode Recordings Using Machine Learning Techniques and Multidimensional...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Extracellular recordings of neuronal spikes are vital for understanding brain activity.
• Current spike classification relies on simplified waveform features, overlooking neuronal diversity.
• Advanced techniques necessitate more sophisticated classification methods.

Purpose of the Study:

• To develop an automatic spike waveform classifier using advanced machine learning.
• To improve the characterization of neuronal discharge patterns and waveform diversity.
• To integrate this classifier into an existing electrophysiological preprocessing pipeline.

Main Methods:

• Utilized machine learning techniques including Uniform Manifold Approximation and Projection (UMAP), Gaussian Mixture Model (GMM), and Random Forest (RF).
• Employed all voltage samples of each waveform for multi-dimensional analysis.
• Trained and tested a Random Forest model on human visual cortex electrophysiological recordings.

Main Results:

• Achieved high performance scores ([Formula: see text]) with the Random Forest classifier.
• Demonstrated improved characterization of waveform clusters compared to a standard toolbox.
• Identified a third distinct group of waveforms, surpassing the binary broad/narrow classification.

Conclusions:

• The developed machine learning classifier offers a more nuanced analysis of neuronal spike waveforms.
• This method enhances the understanding of cortical neuron diversity and discharge patterns.
• The classifier provides a significant improvement over existing waveform analysis techniques.