KIASORT
Knowledge-Integrated Automated Spike Sorting
Modern high-density neural recordings demand spike sorting algorithms that can handle diverse probe geometries and complex, neuron-specific drift, yet existing methods often rely on rigid geometric assumptions and one-dimensional drift models. KIASORT introduces a geometry-free approach for per-neuron drift tracking, training channel-specific classifiers in a hybrid linear–nonlinear embedding space.
Key Features
Geometry-Free Tracking
First-ever per-neuron tracking system that abandons rigid geometric assumptions, handling each neuron's unique drift pattern independently.
Hybrid Dimensionality Reduction
Combines linear PCA with nonlinear UMAP to preserve both global and local geometry of spike waveforms, avoiding linear variability assumptions.
Future-Ready Design
Compatible with next-generation flexible probes from Neuralink, Synchron, and Paradromics, handling complex non-rigid deformations.
Modular Architecture
Three-module design enables real-time processing for closed-loop experiments and brain-computer interfaces.
Integrated Platform
Unified GUI combines automated sorting and manual curation, streamlining workflow from raw data to final results.
Accessibility
Achieves real-time performance using only standard CPUs—not feasible with comparable algorithms. No specialized hardware required.
Key Innovations
Geometry-Free Per-Neuron Tracking
Through sophisticated biophysical simulation, we document that even sub-micron probe displacements (< 1 μm) cause neuron-specific waveform changes that existing drift correction methods cannot handle. KIASORT's per-neuron tracking algorithm addresses each neuron's unique drift pattern without geometric constraints.
Novel Hybrid Embedding Space
While existing methods rely solely on linear embeddings, we combine PCA with UMAP to preserve both global and local geometry. This approach avoids the core assumption of linear variability, leading to more accurate clustering of neurons with nonlinear and temporally changing waveforms.
Real-Time Modular Processing
KIASORT's three-module design (sampling, clustering/training, and sorting) uniquely enables online implementations. Once trained, the sorting module processes incoming streams in real time using pre-trained classifiers—essential for closed-loop experiments and clinical applications.
KIASORT in Action
Hover to see after
Denoising Capability
Advanced noise reduction preserving neural signal integrity. Our sophisticated denoising algorithm removes artifacts while maintaining the crucial features of neural waveforms.
Neuron-Specific Drift Tracking
Visualizing how KIASORT handles unique drift patterns for individual neurons. Each neuron's movement is tracked independently, ensuring accurate spike sorting even with complex probe movements.
Real Data Example
Showing performance of KIASORT on different high density probes. KIASORT consistently identifies 5-15% more high-quality units than current state-of-the-art algorithms across various probe geometries and recording conditions.
Integrated GUI
Complete workflow from raw data to curated results in one interface. The unified platform combines automated sorting with manual curation tools for optimal results.
Get KIASORT
Open Source on GitHub
Access the complete source code, contribute to development, and join our growing community of neuroscientists and developers.
View on GitHubLearn KIASORT
+Installation Guide
Step-by-step instructions for installing KIASORT with all dependencies.
Download PDFVideo Tutorial
Get in Touch
Contact Us
Have questions about KIASORT or want to collaborate? We'd love to hear from you.
Kianooshbb@gmail.com