Tracking single units in chronic, large scale, neural recordings for brain machine interface applications.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Tracking neuronal unit identity is crucial for understanding population coding and ensuring stable neural decoding in brain-machine interfaces (BMIs).
• Variability in neural signals necessitates frequent manual calibration of BMIs, posing a challenge for chronic use and clinical applications.

Purpose of the Study:

• To develop and validate an efficient, autonomous algorithm for tracking single-unit stability across multiple recording sessions.
• To reduce the reliance on manual calibration in BMIs by automating the identification of stable neural units.

Main Methods:

• An algorithm was developed to build a database of features from average spike waveforms and firing patterns.
• A classification approach was used to determine if units recorded on different days belong to the same single-unit.
• Performance was assessed using human expert judgment, measuring accuracy and execution time.

Main Results:

• The algorithm achieved approximately 90% classification accuracy with an average processing time of 12 seconds per channel.
• 77% of automatically tracked single-units matched those identified by human experts.
• A trade-off between accuracy and execution time was observed with increasing data volumes.

Conclusions:

• Automated unit tracking can be performed with high accuracy over months of recordings, significantly streamlining BMI calibration.
• This technology can improve the reliability of BMIs and accelerate their clinical deployment.
• Findings support the study of population coding during learning and enhance BMI system performance.