Optimizing the automatic selection of spike detection thresholds using a multiple of the noise level

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Thresholding is a common, computationally simple spike detection method for neural signal processors.
• Automatic threshold selection is essential for high-channel count data acquisition systems.
• Estimating noise levels and using a multiplier is a simple approach for automatic thresholding.

Purpose of the Study:

• To analyze the effectiveness of different noise estimation operators for automatic threshold selection in neural spike detection.
• To identify optimal multipliers for noise estimation methods in brain-machine interface applications.
• To compare the performance of various noise estimation techniques for neural waveform analysis.

Main Methods:

• Four distinct operators were employed to estimate noise levels in neural waveforms.
• Thresholds for spike detection were set based on these noise estimates.
• An optimization framework was used to identify the optimal multiplier for each noise measure.
• Performance was evaluated using a metric suitable for brain-machine interface applications.

Main Results:

• The root-mean-square operator was found to be the least advantageous for setting detection thresholds.
• Specific optimal multipliers were identified for each noise estimation method.
• The analysis provides a framework for optimizing automatic threshold selection.
• Different noise estimation techniques yielded varying degrees of effectiveness.

Conclusions:

• The choice of noise estimation operator significantly impacts the performance of automatic threshold selection for neural spike detection.
• The root-mean-square method is suboptimal for this application compared to other analyzed methods.
• The presented optimization framework can guide researchers in selecting appropriate methods for unsupervised spike detection.
• Further development of unsupervised methods can benefit from this analysis.