Real-Time Readout of Large-Scale Unsorted Neural Ensemble Place Codes

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Understanding brain function relies on decoding neural activity.
• Real-time analysis of large-scale neural ensembles is crucial for closed-loop experiments.
• Existing methods for population decoding are often computationally intensive.

Purpose of the Study:

• To develop a graphics processing unit (GPU)-powered system for ultrafast population decoding of spatial positions from rodent neural activity.
• To enable real-time reconstruction of spatial trajectories during behavior and sleep.
• To facilitate closed-loop neuroscience experiments by providing rapid neural data analysis.

Main Methods:

• Developed a GPU-accelerated population-decoding system for analyzing spatiotemporal spiking patterns.
• Compared the performance against an optimized central processing unit (CPU) implementation.
• Implemented real-time parallel shuffling for statistical significance assessment of decoded events.

Main Results:

• Achieved a 20- to 50-fold increase in decoding speed compared to CPU methods.
• Demonstrated real-time decoding speeds (milliseconds per spike) scalable to thousands of channels.
• Enabled rapid assessment of memory replay candidates during quiet wakefulness and sleep.

Conclusions:

• The GPU-powered system offers ultrafast population decoding for neuroscience research.
• This toolkit significantly enhances the capability for real-time analysis of neural ensemble activity.
• Facilitates advanced closed-loop experiments, including the study of memory replay.