Enhancing generalizability in classification of peripheral neural recordings with graph neural network
Rui Qi Ji, Mehdy Dousty, Ryan G. L. Koh, Ervin Sejdić
Abstract
The peripheral nervous system plays a crucial role in facilitating communication between biological systems. However, decoding neural signals from peripheral nerve recordings remains a challenge due to their complex spatiotemporal patterns. In this study, we propose a graph-based learning approach to more effectively capture temporal and spatial information for classifying neural signal patterns.
Introduction
The nervous system is a fundamental component of biological function, serving as the primary communication network within the body [1]. It is responsible for transmitting electrical and chemical signals that regulate movement, sensation, and autonomic processes [2,3].
Materials and methods
Neural recordings were previously collected from nine Long-Evan rats from the sciatic nerve using a 56-multi-contact nerve cuff electrode were used [11]. In this study, we excluded one rat due to an issue with the degradation of the plantarflexion signal, resulting in a dataset comprising of eight rats.
Results
Table 2 presents the classification accuracies (%) and macro-averaged F1 score for each individual rat when used as a test set, as well as the mean and standard deviation across all rats. The first row shows the performance of the baseline CNN model reimplemented from the previous study [12] and the subsequent rows show the graph-based model proposed in this work.
Discussion
The proposed approach in this work showed the significance and effectiveness of using graphs to encode information from neural recordings. The results demonstrated that the proposed graph-based learning approach outperforms conventional CNNs in classifying peripheral neural signals, both in across-subject generalization and within-subject evaluations.
Conclusion
We explored a graph-based approach for classifying three afferent activities using neural recordings obtained from Long-Evan rats with a 56-channel nerve cuff electrode. The GNN model effectively captured temporal patterns through nodal features, and by constructing weighted graph adjacency matrices with geodesic distance.
Citation: Ji RQ, Dousty M, Koh RGL, Sejdić E (2026) Enhancing generalizability in classification of peripheral neural recordings with graph neural network. PLoS One 21(4): e0345204. https://doi.org/10.1371/journal.pone.0345204
Editor: Luca Citi, University of Essex, UNITED KINGDOM OF GREAT BRITAIN AND NORTHERN IRELAND
Received: August 29, 2025; Accepted: March 3, 2026; Published: April 17, 2026
Copyright: © 2026 Ji et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The data used in this study were from a publicly available dataset available at Borealis, U of T Dataverse: https://doi.org/10.5683/SP3/JRZDDR.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
