Spike Sorting of Non-Stationary Data in Successive Intervals Based on Dirichlet Process Mixtures

Area of Science:

• Computational Neuroscience
• Machine Learning
• Signal Processing

Background:

• Spike sorting is crucial for analyzing neural activity.
• Tracking non-stationary neural data presents significant challenges.
• Existing methods may struggle with dynamic changes in neural signals.

Purpose of the Study:

• To develop an automatic spike sorting and tracking method for non-stationary data.
• To improve the accuracy and robustness of neural data analysis.
• To address limitations of standard Dirichlet Process Mixture (DPM) models.

Main Methods:

• Utilizes a Dirichlet Process Mixture (DPM) model adapted for sequential data.
• Divides data into intervals and applies mixture models to individual frames.
• Leverages information from previous frames to estimate current cluster parameters.
• Employs non-informative priors for parameter estimation.

Main Results:

• The proposed method successfully tracks variations in cluster size, shape, and location.
• It effectively detects the appearance and disappearance of neural clusters.
• Demonstrates superior performance compared to standard DPM on simulated and real neural data.
• Validated using principal component analysis (PC1-PC2) and applicable to other feature extraction methods.

Conclusions:

• The novel DPM-based method offers enhanced capabilities for spike sorting and tracking.
• It provides a robust solution for analyzing non-stationary neural data.
• The approach shows significant improvements over traditional methods in dynamic neural signal analysis.