A Proof-of-Concept Numerical Ising Machine for Neural Spike Localization

Area of Science:

• Computational Neuroscience
• Signal Processing
• Optimization Techniques

Background:

• Spike localization, identifying neuron physical locations from recorded signals, can improve spike sorting accuracy.
• Current methods face challenges with high channel counts and data volume in neural monitoring.

Purpose of the Study:

• To propose and evaluate a new heuristic method for spike localization.
• To formulate spike localization as a Quadratic Unconstrained Binary Optimization (QUBO) problem.
• To assess the feasibility of using an Ising machine solver for this optimization problem.

Main Methods:

• Formulated the spike localization problem as a QUBO.
• Developed a MATLAB solver simulating an Ising machine to solve the QUBO.
• Evaluated the method on a 2D toy problem with varying distances between electrodes and neuron location.

Main Results:

• Accurate neuron identification was achieved when the neuron was close to the electrodes.
• Localization accuracy decreased to 87.5% and 75% as the search region distance increased.
• Demonstrated the first QUBO formulation for spike localization and its heuristic solution via an Ising machine.

Conclusions:

• The proposed method successfully formulates and heuristically solves the spike localization problem using an Ising machine.
• This approach holds potential for reducing data volume and power consumption in high-channel-count implantable neural monitoring systems.
• Accurate spike localization is feasible, especially for neurons in proximity to recording electrodes.