Computationally inexpensive enhanced growing neural gas algorithm for real-time adaptive neural spike clustering

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Real-time closed-loop neural feedback requires rapid analysis of brain activity.
• Existing spike clustering algorithms are computationally intensive and not optimized for real-time processing.

Purpose of the Study:

• To develop a computationally lightweight and memory-conserving spike clustering algorithm for real-time neural signal processing.
• To enable adaptation to changing electrophysiological environments and classification of both pre-recorded and streaming action potentials.

Main Methods:

• Developed the enhanced growing neural gas (EGNG) algorithm, utilizing a minimal number of nodes and edges.
• Minimized memory requirements by avoiding data retention and focusing on computationally inexpensive Euclidean distance calculations.
• Implemented EGNG in hardware using a Field Programmable Gate Array (FPGA) for real-time performance evaluation.

Main Results:

• EGNG demonstrated efficient classification of neural spikes with minimal computational resources.
• Hardware implementation achieved a worst-case clustering latency of 3.10 µs, processing over 322,580 neural spikes per second.
• The algorithm successfully classified both synthetic and pre-recorded neural spikes, including streaming data.

Conclusions:

• The EGNG algorithm offers a viable solution for real-time neural spike classification.
• Its low computational and memory demands make it suitable for front-end processing in miniaturized, closed-loop neural feedback systems.