A morphology-based feature set for automated Amyotrophic Lateral Sclerosis diagnosis on surface electromyography. (January 2023)

• Record Type: Journal Article
• Authors: Antunes, Margarida; Folgado, Duarte; Barandas, Marília; Carreiro, André; Quintão, Carla; de Carvalho, Mamede; Gamboa, Hugo
• Is Part Of: Biomedical signal processing and control. Volume 79(2023)Part 1
• Publication Date: 2023-01
• Subjects: Amyotrophic Lateral Sclerosis -- Surface electromyography -- Time series -- Signal processing -- Feature selection -- Machine learning; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104011
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Amyotrophic Lateral Sclerosis (ALS) is a fast-progressing disease with no cure. Nowadays, needle electromyography (nEMG) is the standard practice for electrodiagnosis of ALS. Surface electromyography (sEMG) is emerging as a more practical and less painful alternative to nEMG but still has analytical and technical challenges. The objective of this work was to study the feasibility of using a set of morphological features extracted from sEMG to support a machine learning pipeline for ALS diagnosis. We developed a novel feature set to characterize sEMG based on quantitative measurements to surface representation of Motor Unit Action Potentials. We conducted several experiments to study the relevance of the proposed feature set either individually or combined with conventional feature sets from temporal, statistical, spectral, and fractal domains. We validated the proposed machine learning pipeline on a dataset with sEMG upper limb muscle data from 17 ALS patients and 24 control subjects. The results support the utility of the proposed feature set, achieving an F 1 score of (81.9 ± 5.7) for the onset classification approach and (83.6 ± 6.9) for the subject classification approach, solely relying on features extracted from the proposed feature set in the right first dorsal interosseous muscle. We concluded that introducing the proposed feature set is relevant for automated ALS diagnosis since it increased the classifier performance during our experiments. The proposed feature set might also help design more interpretable classifiers as the features give additional information related to the nature of the disease, being inspired by the clinical interpretation of sEMG. Highlights: We proposed a set of novel features for sEMG based on signal morphology. A validation study with sEMG with 17 ALS patients and 24 control subjects. A practical data collection setup with promising results for ALS diagnosis. Features based on sEMG morphology might improve the classifiers' interpretability. The features might have potential as novel biomarkers of ALS disease progression.