RT-Sort: An action potential propagation-based algorithm for real time spike detection and sorting with millisecond...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Bioengineering

Background:

• High-density multi-electrode devices enable detection of individual neuron action potentials.
• Traditional spike sorting occurs post-recording, limiting real-time applications.
• Real-time analysis is crucial for advanced neuroscience experiments.

Purpose of the Study:

• To develop a spike sorting algorithm capable of real-time analysis during neural recordings.
• To enable closed-loop experiments with minimal latency.

Main Methods:

• Utilized unique action potential propagation patterns along axons.
• Employed high-fidelity sequential activations on adjacent electrodes.
• Developed a convolutional neural network-based spike detection algorithm integrated into RT-Sort.

Main Results:

• RT-Sort achieves real-time spike sorting within 7.5ms ± 1.5ms of waveform trough.
• Demonstrated functionality on Multi-Electrode Arrays and Neuropixels probes.
• Enabled closed-loop experiments with latencies comparable to synaptic delays.

Conclusions:

• RT-Sort provides true real-time spike sorting capabilities.
• This advancement significantly enhances the potential for dynamic neuroscience research.
• RT-Sort is adaptable to various electrophysiological recording hardware.