Performance comparison of extracellular spike sorting algorithms for single-channel recordings

Area of Science:

• Computational Neuroscience
• Signal Processing

Background:

• Accurate classification of action potentials from extracellular recordings is crucial for understanding neuronal behavior.
• Numerous spike sorting algorithms exist, but comparative analyses are lacking.

Purpose of the Study:

• To compare the performance of three widely-used spike sorting algorithms: WaveClus, KlustaKwik, and OSort.
• To evaluate the impact of parameter optimization on algorithm accuracy.

Main Methods:

• Utilized 112 artificial extracellular recordings with varying neuron counts and noise levels.
• Employed an optimization technique based on Adjusted Mutual Information to determine near-optimal parameter settings.
• Compared algorithm performance using default versus optimized parameters.

Main Results:

• All three algorithms performed significantly better with optimized parameters (p<0.01).
• WaveClus achieved the highest accuracy, outperforming others for 60% of signals.
• OSort was significantly faster (nearly 5x) but less accurate than WaveClus in moderate noise (0.15-0.30).
• KlustaKwik performed comparably to WaveClus in low noise (0.00-0.15) but was less accurate otherwise.

Conclusions:

• No single spike sorting algorithm is universally optimal.
• Algorithm performance is highly dependent on parameter tuning and specific signal characteristics.
• Optimized parameters enhance spike sorting accuracy across different algorithms.