KIASORT: Knowledge-Integrated Automated Spike Sorting for Geometry-Free Neuron Tracking.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Modern neural recordings require advanced spike sorting algorithms.
• Existing methods struggle with diverse probe geometries and complex, neuron-specific signal drift.
• Current drift models often rely on rigid geometric assumptions and limited dimensionality.

Purpose of the Study:

• Introduce KIASORT (Knowledge-Integrated Automated Spike Sorting), a novel geometry-free spike sorting approach.
• Develop a method for accurate per-neuron drift tracking in high-density neural recordings.
• Provide a user-friendly platform for automated and manual spike sorting.

Main Methods:

• KIASORT employs channel-specific classifiers trained in a hybrid linear-nonlinear embedding space.
• It tracks each neuron's drift independently, unconstrained by probe geometry.
• The method captures waveform features missed by conventional linear approaches.

Main Results:

• Biophysical simulations demonstrated that KIASORT corrects neuron-specific waveform distortions caused by probe displacements.
• In ground-truth benchmarks, KIASORT significantly outperformed Kilosort4 in recovering high-quality neural units.
• The algorithm maintained real-time performance on standard CPUs and was validated on primate and mouse data.

Conclusions:

• KIASORT offers a robust and accurate solution for spike sorting with complex neural data.
• Its geometry-free, per-neuron drift tracking overcomes limitations of existing methods.
• The integrated graphical interface provides a complete and user-friendly spike sorting platform.