MCWs: MicroWire Sorter

An open-source, modular framework for automated spike sorting in human intracerebral recordings.

Overview

MCWs (MicroWire sorter) is a modular, open-source spike sorting framework engineered specifically for human intracerebral recordings in clinical settings. Developed in collaboration with the Medical College of Wisconsin ReyLab, MCWs addresses the unique challenges of processing neural data from epilepsy patients undergoing invasive monitoring.

The Problem

Spike sorting — the process of detecting, extracting, and classifying neural action potentials from extracellular recordings — is essential for studying the neural basis of human cognition and for advancing epilepsy diagnostics. However, existing tools are primarily designed for animal recordings and struggle with:

  • Clinical noise environments — Hospital settings introduce artifacts not present in controlled lab recordings
  • Human-specific waveforms — Human single-unit waveforms differ from animal data in shape, amplitude, and variability
  • Scale — Modern microwire bundles generate massive datasets requiring efficient, automated processing

Our Approach

MCWs provides an end-to-end pipeline for clinical spike sorting:

1. Preprocessing & Artifact Rejection

  • Custom filtering optimized for human microwire recordings (bandpass: 300 Hz – 3 kHz)
  • Novel artifact detection methods designed for hospital-based electrophysiology environments
  • Integration with SpikeInterface for flexible data format support

2. Spike Detection

  • Wavelet-based detection with adaptive thresholding
  • Handles overlapping spikes and low signal-to-noise ratios common in clinical data

3. Feature Extraction

  • Haar wavelet decomposition for capturing waveform morphology
  • Principal Component Analysis (PCA) for dimensionality reduction
  • Hybrid feature pipelines selectable per recording quality

4. Clustering

  • Super Paramagnetic Clustering (SPC) for robust neuron identification
  • Automatic cluster count determination without manual intervention
  • Quality metrics for cluster validation and merge decisions

Technical Architecture

Raw Recording → Bandpass Filter → Artifact Rejection → Spike Detection
    → Feature Extraction (Wavelets / PCA) → SPC Clustering → Sorted Units

Infrastructure

MCWs is deployed on a high-performance computing environment at MCW:

  • 40-core systems for parallel processing of multi-channel recordings
  • 250 TB storage for large-scale neural data archives
  • Automated quality control and reproducible analysis pipelines

Impact

MCWs enables researchers to investigate:

  • Episodic memory — How single neurons encode and retrieve memories
  • Seizure onset localization — Identifying seizure focus through single-neuron firing patterns
  • Local field potential biomarkers — Discovering neural signatures for epilepsy diagnosis

Status: Preprint available on bioRxiv (first co-author) Read the preprint →