A Particle Swarm Optimized Independence Estimator for Blind Source Separation of Neurophysiological Time Series

Area of Science:

• Neuroscience
• Neuroengineering
• Signal Processing

Background:

• Decomposing neurophysiological recordings into neural sources is crucial for neuroscience and neuroengineering.
• High-density electrode arrays necessitate advanced blind source separation methods.
• Traditional Independent Component Analysis (ICA) struggles with sparse sources due to implementation inefficiencies.

Purpose of the Study:

• To address the limitations of single non-linear optimization functions in ICA for spiking sources.
• To develop a novel projection-pursuit ICA algorithm tailored for efficient neural source separation.
• To improve the recovery and accurate separation of all spiking sources in neurophysiological signals.

Main Methods:

• Developed a projection-pursuit ICA algorithm utilizing a particle swarm methodology.
• The algorithm adaptively traverses a polynomial family of non-linearities to approximate source asymmetric cumulants.
• Tested on recordings from high-density intramuscular probes.

Main Results:

• The proposed algorithm demonstrates state-of-the-art decomposition performance.
• The particle swarm efficiently identifies optimal contrast non-linearities.
• Successfully overcomes limitations of single non-linear optimization in ICA for spiking sources.

Conclusions:

• The novel projection-pursuit ICA algorithm effectively separates spiking neural sources.
• This method enhances the analysis of neurophysiological data from high-density recordings.
• Offers a robust solution for improving neural source decomposition in neuroscientific research.