Spike sorting with Gaussian mixture models.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Extracellular recordings capture neuronal electrical activity, but isolating individual neuron signals is challenging.
• Spike sorting classifies recorded waveforms to identify distinct neuronal sources.
• Current methods face limitations in accurately assigning spike waveforms to their neurons of origin.

Purpose of the Study:

• To evaluate Gaussian mixture models (GMMs) for feature extraction and clustering in spike sorting.
• To develop and validate a GMM-based framework for improved unsupervised spike sorting.
• To compare the performance of the GMM approach against established spike sorting techniques.

Main Methods:

• Feature extraction using principal components, wavelets, and GMM fitting parameters.
• Unsupervised clustering of spike waveforms utilizing GMMs.
• Data-driven estimation of cluster properties for robust classification.
• Validation using simulated and real extracellular electrophysiological recordings.

Main Results:

• The proposed GMM-based spike sorting framework demonstrated superior performance compared to existing methods.
• The GMM approach achieved higher accuracy in classifying spike waveforms from both simulated and real data.
• Identified potential advantages of GMM-based feature extraction over traditional principal components analysis.

Conclusions:

• Gaussian mixture models offer a powerful and effective approach for enhancing spike sorting accuracy.
• The developed GMM framework provides a robust solution for identifying individual neuronal signals in complex extracellular recordings.
• The study introduces a user-friendly interface for applying the GMM-based spike sorting algorithm, facilitating broader adoption.