Hybrid knowledge- and data-driven modelling for robust spike detection and sorting in human C-fiber microneurography

2026-03-12 · Source: Machine learning : nature.com subject feeds · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Neuroscience & Neurophysiology · Depth: Advanced, extended

Summary

Researchers have developed a hybrid knowledge- and data-driven computational pipeline for robust spike detection and sorting in human C-fiber microneurography, addressing challenges like single-electrode recordings, waveform variability, and low signal-to-noise ratios. The pipeline combines knowledge-driven steps, such as extracting spike templates from electrically evoked spikes and restricting detection to intervals showing activity-dependent latency shifts, with data-driven machine learning using One-class SVM, SVM, and XGBoost classifiers. A specialized "ground truth" stimulation protocol was created to provide reliable labels for all electrically evoked spikes, enabling precise validation. The approach was benchmarked against Spike2 software, achieving higher F1-scores and reduced false positives, though optimal performance was highly dependent on individual combinations of feature sets and models for each recording. A proof-of-concept application to chemically induced C-fiber activity demonstrated its potential for sensory spike train analysis, marking a significant step towards reliable analysis of pain and itch signaling.

Key takeaway

For AI Scientists developing neural signal processing tools for microneurography, this work highlights the necessity of a hybrid approach that integrates physiological knowledge with data-driven machine learning. You should prioritize dataset-specific optimization of feature sets and classification models, as no universal solution exists. Consider implementing automated sortability criteria based on SNR, template similarity, and waveform drift to enhance reliability and reduce manual intervention in your pipelines.

Key insights

A hybrid pipeline improves C-fiber spike sorting in microneurography by combining physiological knowledge with machine learning for robust detection.

Principles

• Activity-dependent latency shifts constrain spike detection.
• No single ML model or feature set is universally optimal.
• Template similarity and fiber count predict sortability.

Method

The pipeline extracts spike templates from electrically evoked spikes, identifies activity-dependent latency shifts to constrain detection, and then uses supervised ML (One-class SVM, SVM, XGBoost) with various feature sets (SPDF, SPDF_FV3, W_raw) for classification.

In practice

• Use a specialized "ground truth" protocol for reliable spike labeling.
• Evaluate SNR and template similarity before sorting.
• Optimize feature set and ML model per recording.

Topics

• Spike Sorting
• Microneurography
• C-fiber Nociceptors
• Machine Learning Classification
• Activity-Dependent Conduction Velocity

Code references

• Digital-C-Fiber/SpikeSortingForSpikingPatterns

Best for: AI Scientist, Machine Learning Engineer, Data Scientist, Research Scientist

Related on AIssential

• A multimodal approach for visualizing and identifying electrophysiological cell types in vivo — PhysMAP integrates multiple electrophysiological modalities to accurately identify neuronal cell types and detect batch effects.
• A neurosymbolic Approach with Epistemic Deep Learning for Hierarchical Image Classification — A neurosymbolic framework enhances hierarchical image classification with epistemic uncertainty and logical consistency.
• Reducing the Complexity of Deep Learning Models for EEG Analysis on Wearable Devices — Parameter quantization and electrode reduction can significantly reduce DNN complexity for wearable EEG analysis with minimal accuracy loss.
• Zyphra Releases ZUNA: A 380M-Parameter BCI Foundation Model for EEG Data, Advancing Noninvasive Thought-to-Text Development — ZUNA is a 380M-parameter BCI foundation model for EEG, using 4D RoPE and diffusion for robust signal reconstruction.
• Beyond Augmentation: Score-Guided Pathological Prior for EEG-based Depression Detection — A score-guided classification framework improves EEG-based depression detection by modeling pathological priors without data augmentation.
• Attention-Gated Convolutional Networks for Scanner-Agnostic Quality Assessment — A hybrid CNN-Attention model effectively assesses MRI quality, generalizing across diverse scanners without retraining.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.