KIASORT

KIASORT implements a geometry-free spike sorting algorithm that tracks neuron-specific drift in high-density neural recordings by training channel-specific classifiers in a hybrid linear-nonlinear embedding space.

Key Features:

• Geometry-free drift tracking: KIASORT tracks each neuron independently without constraints from probe layout or rigid geometric assumptions.
• Hybrid linear-nonlinear embedding: Channel-specific classifiers are trained in a hybrid linear-nonlinear embedding space to capture waveform features often missed by conventional linear methods.
• Real-time performance: KIASORT maintains real-time performance on standard CPUs.
• Automated sorting with manual curation: The platform combines automated sorting with manual curation capabilities.

Scientific Applications:

• High-density neural recordings with diverse probe geometries: KIASORT handles recordings from probes with varied geometries that challenge methods relying on rigid geometric assumptions.
• Neuron-specific drift correction: KIASORT outperforms existing methods such as Kilosort4 in recovering high-quality units in ground-truth benchmarks with heterogeneous, neuron-specific drift, including sub-micron probe displacements that induce neuron-specific waveform distortions.
• Cross-species neural data analysis: KIASORT has been validated on both primate and mouse neural recordings.

Methodology:

KIASORT trains channel-specific classifiers in a hybrid linear-nonlinear embedding space, then sorts spikes by independently tracking each neuron without geometric constraints. Performance was evaluated using biophysical simulations and ground-truth benchmarks with heterogeneous, neuron-specific drift.