Neuromuscular disorders detection through time-frequency analysis and classification of multi-muscular EMG signals using Hilbert-Huang transform. (January 2022)

• Record Type: Journal Article
• Authors: Torres-Castillo, Jonathan R.; López-López, Carlos Omar; Padilla-Castañeda, Miguel A.
• Is Part Of: Biomedical signal processing and control. Volume 71(2022)Part A
• Publication Date: 2022-01
• Subjects: Non-stationary processing of biosignals -- Time-frequency analysis -- Adaptive signal decomposition -- 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.2021.103037
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Electromyographic (EMG) signal analysis plays a vital role in diagnosing neuromuscular disorders (NMD). It is based on the clinician's experience in interpreting the signal's shape and acoustic properties. For accurate detection of these disorders, developing new techniques to analyze these signals comprehensively has increased. This paper presents a machine learning strategy (ML) to classify EMG signals to automatically detecting the presence of neuropathy, myopathy, or absence of disease efficiently. A database of 938 signals acquired from different muscles divided symmetrically into these three classes was used. The method decomposes each signal into amplitude or frequency modulated sub-bands and extracts from them time-frequency features using the Hilbert Transform. Non-parametric statistical analysis and Uncorrelated Linear Discriminant Analysis (ULDA) were used for feature selection and data's dimensionality reduction. Three different techniques of ML were used in the classification; LDA, TREE, and KNN. Five decomposition methods were evaluated: empirical mode decomposition (EMD), ensemble EMD (EEMD), complementary EEMD (CEEMD), empirical wavelet transform (EWT), and variational mode decomposition (VMD). The best results were achieved by using EEMD with the KNN with an accuracy of 99.5%. A sensitivity of 99.6% to the neuropathic and 98.8% for myopathic. A specificity of 99.2% and a positive predictive value of 99.6%. This study has the highest classification performance with such variability and extensive data, outperforming the state-of-the-art neuromuscular disorders classification. Consequently, the proposed methodology adequately interprets the information extracted from the quantitative time-frequency analysis, showing its validity to support the clinician in detecting and diagnosing NMD highly efficiently.