Evaluation study of compressed sensing for neural spike recordings

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Modern neural recorders generate high data rates (16-20 Mbit/s) from numerous parallel channels.
• Low-power design is crucial for implantable neural recorders, necessitating data compression before wireless transmission.
• Compressing neural data enables flexible data analysis at the receiver.

Purpose of the Study:

• To evaluate the performance of compressed sensing (CS) for neural action potential (spike) signal compression.
• To assess the feasibility of using CS for reducing data rates in neural recording systems.
• To determine the applicability of CS for both synthesized and recorded neural data.

Main Methods:

• Utilized compressed sensing (CS) techniques in MATLAB for neural spike signal compression.
• Employed a 6-level Daubechies-8 wavelet decomposition matrix and two learned dictionary matrices as CS dictionaries.
• Evaluated compression efficacy using synthesized and recorded neural datasets.
• Assessed the performance of CS using the OSort spike sorting program.

Main Results:

• Compressed sensing (CS) demonstrated successful compression of low-noise synthesized neural spike signals.
• Achieved a compression rate of 2.05 for synthesized neural spike data.
• Performance on recorded neural spike signals was not explicitly detailed but implied limitations.

Conclusions:

• Compressed sensing (CS) is effective for compressing low-noise synthesized neural spike signals.
• CS is not recommended for general compression of neural spike signals due to limitations.
• Further research may be needed to optimize CS for diverse neural recording conditions.