Intelligent Technique for Neuromuscular Disorders Prediction Based on Stockwell Transform

Abstract

Electromyography (EMG) signals, which reflect the electrical activity of skeletal muscles, serve as a fundamental tool in diagnosing neuromuscular disorders. Yet, due to their inherently nonstationary characteristics, extracting meaningful features from these signals remains a persistent challenge in the field. To address this, the present study introduces a hybrid methodological framework. Initially, EMG signals are transformed into time–frequency representations using the Stockwell transform (ST), effectively converting complex signal data into image-like formats amenable to advanced analysis. Subsequently, the framework leverages the power of five pretrained deep convolutional neural networks (CNNs)—namely, VGG16, ResNet50, DenseNet201, InceptionV3, and InceptionResNetV2—as automated feature extractors. This approach reduces dependence on traditional handcrafted features and exploits the capacity of CNNs to uncover intricate signal patterns. The investigation utilizes the publicly available EMGLAB database, encompassing EMG data from individuals diagnosed with amyotrophic lateral sclerosis, myopathy, and healthy controls. The signals are segmented and transformed into Stockwell-based images prior to feature extraction. For classification, features derived from the CNNs are evaluated using conventional machine learning algorithms, including support vector machine (SVM), random forest, knearest neighbor, and naïve Bayes. Empirical results reveal that the combination of VGG16 and SVM produces the most accurate classification, with achieved accuracies ranging between 97.0% and 98.5% across five distinct classification tasks. These findings underscore the efficacy and potential of the ST combined with CNN-based strategies for robust and accurate classification of neuromuscular disorders, establishing a valuable benchmark for future EMG research. 

Received: 31 August 2025 | Revised: 7 January 2026 | Accepted: 27 February 2026

Conflicts of Interest 

The authors declare that they have no conflicts of interest to this work. 

Data Availability Statement 

The data that support the findings of this study are openly available in dataset N2001 at http://www.emglab.net. 

Author Contribution Statement

Nahla F. Abdel-Maboud: Conceptualization, Methodology, Software, Investigation, Resources, Data curation, Writing – original draft, Visualization. Marco Alfonse: Validation, Formal analysis, Investigation, Writing – review & editing, Visualization. Silvia Stoyanova Parusheva: Validation, Writing – review & editing, Visualization, Supervision. Abdel-Badeeh M. Salem: Validation, Writing – review & editing, Visualization, Supervision, Project administration.