Deep neural networks to classify motor unit action potential signals acquired using needle electromyography

Isil Tatlidil; Murat Ekinci; Cavit Boz

doi:10.20883/medical.e1391

Authors

Isil Tatlidil Karadeniz Technical University Faculty of Medicine, Department of Clinical Neurophysiology, Trabzon, Turkiye https://orcid.org/0000-0002-9666-3647
Murat Ekinci b- Karadeniz Technical University Engineering Faculty, Department of Computer Engineering, Trabzon, Turkiye https://orcid.org/0000-0001-9326-8425
Cavit Boz c- Karadeniz Technical University Faculty of Medicine Department of Clinical Neurophysiology, Trabzon, Turkiye https://orcid.org/0000-0003-0956-3304

DOI:

https://doi.org/10.20883/medical.e1391

Keywords:

deep learning, convolutional neural network, gate recurrent unit, needle electromyography, neuromuscular disorders

Abstract

Aim. To classify motor unit action potential patterns using a deep learning technique with high accuracy.

Material and methods. A dataset was compiled from three main groups of motor unit action potential patterns, including myopathy, neuropathy, and normal, as assessed by a clinical neurophysiologist during routine clinical assessments. After preprocessing the raw signals in the dataset, a total of 3,152 signal segments from 96 muscles of 26 individuals were divided into training and test sets. Deep learning network models were developed in Python using the Keras API in Jupyter Notebook.

Results. Among the deep learning models, a hybrid deep neural network model with a one-dimensional convolution layer as an input layer and four layers of gate recurrent units (1DCNN-GRU) achieved the highest accuracy rates. Ten-fold cross-validation resulted in a mean accuracy rate of 98.13 ± 1.05%.

Conclusions. Both conventional machine learning models and deep learning models could classify needle EMG patterns that belonged to three neuromuscular disorder groups with high accuracy. However, more clinical studies with larger datasets are needed for validation. In contrast to conventional machine learning techniques, deep learning models could receive signals as input data and automatically extract the required features. Therefore, they could facilitate the real-time implementation of the pattern recognition tasks in the future.

Downloads

Download data is not yet available.

Author Biographies

Isil Tatlidil, Karadeniz Technical University Faculty of Medicine, Department of Clinical Neurophysiology, Trabzon, Turkiye

Isil Tatlidil, MD is a neurologist, clinical neurophysiologist and computer engineer. She graduated from Dokuz Eylul University Medical Faculty, Izmir, Turkey, in 2012. She completed her neurology training at Katip Celebi University, Izmir, Turkey, in 2016. In 2015, she passed the European Board of Neurology exam. She worked as a neurologist in Malatya Research and Training Hospital. She completed her clinical neurophysiology training at Karadeniz Technical University Medical Faculty in 2023. She also graduated from the computer engineering program of Engineering Faculty of Karadeniz Technical University in 2023. Currently, she works in Agri Research and Training Hospital as a clinical neurophysiologist.
Murat Ekinci, b- Karadeniz Technical University Engineering Faculty, Department of Computer Engineering, Trabzon, Turkiye

Prof. Dr. Murat Ekinci is a professor of computer engineering at Karadeniz Technical University in Trabzon, Turkey. He graduated from the electronic engineering program of the Engineering Faculty of Karadeniz Technical University in 1990. He completed his master’s degree in the Computer Science program of the Science Faculty of Karadeniz Technical University in 1993. Then, he completed his PhD at the computer science program of the Engineering Faculty of the University of Bristol in 1997. He became assistant professor at Karadeniz Technical University in 1998. After being promoted to associate professor in 2010, he became a full professor of computer engineering in the Engineering Faculty of Karadeniz Technical University. He has 198 publications, and his H index is 11 in WOS.
Cavit Boz, c- Karadeniz Technical University Faculty of Medicine Department of Clinical Neurophysiology, Trabzon, Turkiye

Prof. Dr. Cavit Boz is an MS specialist, clinical neurophysiologist and professor of neurology at Karadeniz Technical University in Trabzon, Turkey. He graduated from Karadeniz Technical University Medical Faculty in 1994. He completed his neurology training at Karadeniz Technical University in 2000 and continued working as a neurologist at Karadeniz Technical University. He became a certified clinical neurophysiologist in 2011. He was a clinical fellow in the MS clinic of the University of British Columbia in Vancouver, Canada, in 2005 and again in 2011. After being promoted to associate professor in 2005, he became a full professor in 2010. He has 376 publications, and his H index is 34 in WOS.

References

De Jonge S, Potters W V, Verhamme C. Artificial intelligence for automatic classification of needle EMG signals: A scoping review. Clin Neurophysiol. 2024 Mar;159:41–55. https://doi.org/10.1016/j.clinph.2023.12.134.

Rubin DI. Needle electromyography: Basic concepts. Handb Clin Neurol. 2019 Jan 1;160:243–56.

Preston DC, Shapiro BE. Electromyography and Neuromuscular Disorders. 4th ed. Philadelphia: Elsevier; 2021. 134–259 p.

Kendall R, Werner RA. Interrater reliability of the needle examination in lumbosacral radiculopathy. Muscle Nerve. 2006 Aug;34(2):238–41. https://doi.org/10.1002/mus.20554.

Shen C, Nguyen D, Zhou Z, Jiang SB, Dong B, Jia X. An introduction to deep learning in medical physics: advantages, potential, and challenges. Phys Med Biol. 2020 Mar 3;65(5):05TR01. https://doi.org/10.1088/1361-6560/ab6f51.

Choi RY, Coyner AS, Kalpathy-Cramer J, Chiang MF, Campbell JP. Introduction to machine learning, neural networks, and deep learning. Transl Vis Sci Technol. 2020 Feb 27;9(2):14. https://doi.org/10.1167/tvst.9.2.14.

Caiafa CF, Sun Z, Tanaka T, Marti-Puig P, Solé-Casals J. Special Issue ‘Machine Learning Methods for Biomedical Data Analysis’. Sensors (Basel). 2023 Nov 24;23(23). https://doi.org/10.3390/s23239377.

Yoo J, Yoo I, Youn I, Kim SM, Yu R, et al. Residual one-dimensional convolutional neural network for neuromuscular disorder classification from needle electromyography signals with explainability. Comput Methods Programs Biomed. 2022 Nov;226:107079. https://doi.org/10.1016/j.cmpb.2022.107079.

Faust O, Hagiwara Y, Hong TJ, Lih OS, Acharya UR. Deep learning for healthcare applications based on physiological signals: A review. Comput Methods Programs Biomed. 2018 Jul;161:1–13. https://doi.org/10.1016/j.cmpb.2018.04.005.

Nodera H, Osaki Y, Yamazaki H, Mori A, Izumi Y, Kaji R. Deep learning for waveform identification of resting needle electromyography signals. Clin Neurophysiol. 2019 May;130(5):617–23. https://doi.org/10.1016/j.clinph.2019.01.024.

Nam S, Sohn MK, Kim HA, Kong HJ, Jung IY. Development of artificial intelligence to support needle electromyography diagnostic analysis. Healthc Inform Res. 2019;25(2):131. https://doi.org/10.4258/hir.2019.25.2.131.

Sengur A, Akbulut Y, Guo Y, Bajaj V. Classification of amyotrophic lateral sclerosis disease based on convolutional neural network and reinforcement sample learning algorithm. Health Inf Sci Syst. 2017 Dec 30;5(1):9. https://doi.org/10.1007/s13755-017-0029-6.

Zhang Z, He C, Yang K. A novel surface electromyographic signal-based hand gesture prediction using a recurrent neural network. Sensors. 2020 Jul 17;20(14):3994. https://doi.org/10.3390/s20143994.

Das S, Tariq A, Santos T, Kantareddy SS, Banerjee I. Recurrent neural networks (RNNs): Architectures, training tricks, and introduction to influential research. Newyork: Humana; 2023. https://doi.org/ 10.1007/978-1-0716-3195-9_4.

Keleş AD, Turksoy RT, Yucesoy CA. The use of nonnormalized surface EMG and feature inputs for LSTM-based powered ankle prosthesis control algorithm development. Front Neurosci. 2023;17:1158280. https://doi.org/10.3389/fnins.2023.1158280.

Aviles M, Alvarez-Alvarado JM, Robles-Ocampo JB, Sevilla-Camacho PY, Rodríguez-Reséndiz J. Optimizing RNNs for EMG Signal Classification: A novel strategy using grey wolf optimization. bioengineering (Basel). 2024 Jan 13;11(1). https://doi.org/10.3390/bioengineering11010077.

Hubers D, Potters W, Paalvast O, de Jonge S, Doelkahar B, Tannemaat M et al. Artificial intelligence-based classification of motor unit action potentials in real-world needle EMG recordings. Clinical Neurophysiology. 2023 Dec;156:220–7. https://doi.org/10.1016/j.clinph.2023.10.008.

Khademi Z, Ebrahimi F, Kordy HM. A review of critical challenges in MI-BCI: From conventional to deep learning methods. J Neurosci Methods. 2023 Jan 1;383:109736. https://doi.org/10.1016/j.jneumeth.

Yang S, Yu X, Zhou Y. LSTM and GRU Neural Network Performance Comparison Study: Taking Yelp review dataset as an example. In: 2020 International Workshop on Electronic Communication and Artificial Intelligence (IWECAI). IEEE; 2020. p. 98–101.

Rivas F, Sierra-Garcia JE, Camara JM. Comparison of LSTM- and GRU-Type RNN networks for attention and meditation prediction on raw EEG data from low-cost headsets. Electronics (Basel). 2025 Feb 12;14(4):707. https://doi.org/10.3390/electronics14040707.

Samanta K, Chatterjee S, Bose R. Neuromuscular disease detection based on feature extraction from time–frequency images of EMG signals employing robust hyperbolic Stockwell transform. Int J Imaging Syst Technol. 2022 Jul 25;32(4):1251–62. https://doi.org/10.1002/ima.22709.

Roy SS, Dey D, Karmakar A, Roy AS, Ashutosh K, Choudhury NR. Detection of abnormal electromyograms employing DWT-based amplitude envelope analysis using Teager energy operator. Int J Biomed Eng Technol. 2022;40(3):224. https://doi.org/10.1504/ijbet.2022.126493

Dubey R, Kumar M, Upadhyay A, Pachori RB. Automated diagnosis of muscle diseases from EMG signals using an empirical mode decomposition-based method. Biomed Signal Process Control. 2022 Jan 1;71:103098. https://doi.org/10.1016/j.bspc.2021.103098

Kamali T, Stashuk DW. Electrophysiological muscle classification using multiple instance learning and unsupervised time and spectral domain analysis. IEEE Trans Biomed Eng. 2018 Nov;65(11):2494–502. https://doi.org/10.1109/TBME.2018.2802200.

Nandedkar SD, Barkhaus PE, Stålberg E V. Motor unit recruitment and firing rate at low force of contraction. Muscle Nerve. 2022 Dec;66(6):750–6. https://doi.org/10.1002/mus.27737.

Masakado Y. Motor unit firing behavior in man. Keio J Med. 1994 Sep;43(3):137–42. https://doi.org/10.2302/kjm.43.137.

Van Putten MJAM, Olbrich S, Arns M. Predicting sex from brain rhythms with deep learning. Sci Rep. 2018 Feb 15;8(1):3069. https://doi.org/10.1038/s41598-018-21495-7.