Advancing Spike Sorting Through Gradient-Based Preprocessing and Nonlinear Reduction With Agglomerative Clustering

Area of Science:

• Computational Neuroscience
• Signal Processing
• Machine Learning

Background:

• Spike sorting is crucial for analyzing neural activity, but current methods lack sufficient accuracy.
• Manual spike sorting is time-consuming and inefficient, especially for visually similar spikes.
• There is a need for highly accurate, automated spike-sorting techniques.

Purpose of the Study:

• To develop a fully automated spike-sorting method with high classification accuracy.
• To improve the reliability of neural data analysis through advanced spike sorting.

Main Methods:

• Employed unsupervised mathematical methods for spike sorting, avoiding the need for training data and reducing computational costs.
• Implemented a two-step methodology: data preprocessing and spike classification.
• Utilized nonlinear transformations, including Uniform Manifold Approximation and Projection (UMAP) and spectral embedding, for optimal feature extraction from spike waveforms, followed by density-based clustering.

Main Results:

• Achieved 100% accuracy for non-overlapping spikes and 99.47% accuracy for overlapping spikes on Dataset1.
• Demonstrated a 12% accuracy improvement on challenging dataset portions.
• Showcased efficacy in unit detection and spike clustering on synthetic data.

Conclusions:

• The proposed method achieves unparalleled accuracy in spike sorting.
• This approach surpasses the performance of current state-of-the-art spike-sorting techniques.