Welcome to MUnitQuest!

Aims and scope

MUnitQuest aims to advance methods for reconstructing motor unit spike trains from high-density surface EMG (HDsEMG) through a community-driven competition. The competition is organized into two challenges (also see the competition timeline)

How does it work

Algorithm Challenge: A competition for motor unit identification methods, with two independent tasks — Isometric and Dynamic — each yielding its own leaderboard. The challenge runs in two sequential phases: a Familiarization Phase (develop and test on publicly available training data) and a Showdown Phase (evaluate on the MUnitQuest data collection). Learn more at the Algorithm Challenge page!

Data Challenge: We believe that the Algorithm Challenge can only be successful with access to data with reliable spike train labels. Therefore, the Data Challenge calls for a community-driven effort to build a diverse, high-quality collection of experimental and simulated HDsEMG datasets with reliably labeled motor unit spike trains. Submissions are assessed by a double-blind expert review panel and released on an open data repository. These datasets form the foundation of the Showdown Phase of the Algorithm Challenge. Learn more at the Data Challenge page!

To participate, register on Codabench then navigate to the Data challenge and Algorithm challenge!

Who is it for

The Algorithm Challenge is intended for algorithm developers working on computational methods that can reconstruct the activity of single neurons from complex mixtures.
The Data Challenge is targeted at experimental researchers who use HDsEMG and simulation scientists developing electrophysiological models that want to apply their methods in a highly relevant field of applied neuromuscular research.

Awards

All teams participating in the Algorithm Challenge receive recognition on a permanent leaderboard (per task), and all contributions from the Data Challenge will be published according to the FAIR (Findable, Accessible, Interoperable, Reusable) principles in an open data repository (mandatory for datasets entering the MUnitQuest data collection, optional otherwise). Furthermore, the top 5 teams per task (Isometric and Dynamic), as well as the top 5 dataset contributions (Data Challenge), will be invited to share their solutions in a special issue of the Journal of Electromyography and Kinesiology.

Motivation and background

Since the development of the concentric needle by Adrian and Bronk nearly 100 years ago, the indirect identification of spinal motor neuron activity from electromyographic signals has shaped our understanding of neuromuscular physiology. Over the last 20 years, the development of blind source separation (BSS) algorithms applied to high-density surface electromyography (HDsEMG) recordings has facilitated the study of motor unit activity in living humans, enhanced the population of detectable motor units, and underscored the technique’s potential in applications such as human-machine interfaces. Despite these achievements, several limitations persist:

Most motor unit identification algorithms are still limited to isometric conditions
Due to the lack of community-accepted benchmarks, tracking methodological progress is difficult
The lack of publicly available datasets and standards for sharing EMG data hinders large-scale performance evaluation through automated pipelines