MCWs (MiCroWire Sorter)

A new framework for automated and reliable spike sorting in human intracerebral recordings

MCWs (MiCroWire Sorter) is a modular pipeline for automated spike sorting of human intracerebral microwire recordings.
It is designed to handle large-scale clinical datasets and to provide transparent, quality-controlled single-unit isolation.

---

Project status

This repository contains a beta version of MCWs (MiCroWire Sorter).

• You may encounter bugs or unexpected behavior.
• If you find an issue, please open a GitHub issue with a short description, error message, and (when possible) a minimal example.
• We are actively developing this pipeline and will continue to update and refine it.

---

Pipeline overview

MCWs implements an end-to-end workflow:

1. Initial Data Loading & Pre-processing

   • Data-driven notch filtering and basic cleaning
   • Channel selection and basic QC

2. Spike Detection

   • Automated thresholding and temporal filtering
   • Candidate spike extraction per channel

3. Artifact Rejection

   • Bundle-level rejection of obvious non-neural events
   • Within-channel rejection using waveform shape and decision logic

4. Feature Extraction

   • Waveform- and time-domain features
   • Distribution-based features (e.g., GMM fits, variability measures)
   • Dimensionality reduction / selection for clustering

5. Clustering (Super Paramagnetic Clustering; SPC)

   • Unsupervised clustering of spike features
   • Automatic identification of putative units

6. Refine Sorting

   • Merge/split operations guided by waveform similarity and firing statistics
   • Task-specific response profiles (e.g., peri-event histograms)

7. Quality Metrics

   • Unit stability and isolation indices
   • Waveform consistency and firing rate characteristics
   • Visual summaries for manual review

---

Key innovations

• Fully modular human-microwire pipeline: every stage (detection, artifact rejection, feature extraction, clustering, refinement) is separable and can be swapped or extended.
• Rich artifact and quality control: two-level artifact rejection and extensive unit quality metrics designed specifically for noisy human intracerebral recordings.

---

Data format requirements

The current implementation expects:

• Neural data in NC5 int16 format, and
• A corresponding NSx.mat file containing all metadata required to read the data
  (e.g., sampling rate, channel map, recording duration, etc.). See "Sample_of_NSx.mat" for an exemplary file.

If your recordings are in a different format, you will need to:

1. Implement a parser that converts your raw data into NC5 int16 files.
2. Create a compatible NSx.mat file with the necessary metadata fields.

Once you have *.nc5 (int16) and the associated NSx.mat, you can use this repository without further changes to the main pipeline.

---