Maschinelles Lernen zur Detektion pathologischer Spontanaktivität in der Elektromyografie anhand von Audiosignalen

Zusammenfassung

Ziel

Die Erkennung pathologischer Spontanaktivität in der Elektromyografie (EMG) ist entscheidend für die Diagnostik neuromuskulärer Erkrankungen. Die vorliegende Studie untersuchte den automatisierten Einsatz von maschinellem Lernen zur Identifikation pathologischer Spontanaktivität aus EMG-Audiodaten.

Methodik

Anonymisierte Audiodaten von 40 Patienten mit Willkür- oder pathologischer Spontanaktivität wurden extrahiert und in kurze Segmente aufgeteilt. Verschiedene Algorithmen wie Random Forest, K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), ein Feedforward-Netzwerk und ein Autoencoder-Netzwerk wurden trainiert und verglichen.

Ergebnisse

Der Random-Forest-Klassifikator erzielte eine Genauigkeit von 0,89. KNN zeigte einen hohen Recall von 1,0 bei 0,72 Genauigkeit, was auf zuverlässige Erkennung hinweist. Gradient-Boosting wies ein ausgewogenes Verhältnis zwischen Precision (0,70) und Recall (0,88) auf. Klassische Methoden lieferten je nach Datenqualität variable Ergebnisse, wobei Gradient Boosting insgesamt am besten abschnitt. Das Feedforward-Netzwerk erreichte in segmentierten 3-Sekunden-Abschnitten eine Precision und einen Recall von 1,0 und zeigte somit die höchste Leistung.

Schlussfolgerung

Maschinelles Lernen kann pathologische Spontanaktivität in EMG-Audiodaten effektiv von Willküraktivität unterscheiden. Insbesondere neuronale Netze bieten großes Potenzial zur Unterstützung von Klinikern in der EMG-Diagnostik bei weiterer Optimierung.

Abstract

Objective

Detecting pathological spontaneous activity in electromyography (EMG) is crucial for diagnosing neuromuscular diseases. This study investigated the automated use of machine learning to identify pathological spontaneous activity from EMG audio data.

Methodology

Anonymized audio data from 40 patients with voluntary or pathological spontaneous activity were extracted and divided into short segments. Various algorithms, including Random Forest, K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), a feedforward network, and an autoencoder network, were trained and compared.

Results

The Random Forest classifier achieved an accuracy of 0.89. KNN demonstrated a high recall of 1.0 with an accuracy of 0.72, indicating reliable detection. Gradient boosting showed a balanced ratio between precision (0.70) and recall (0.88). Classical methods produced variable results depending on data quality, with gradient boosting performing the best overall. The feedforward network achieved a precision and recall of 1.0 in segmented 3-second intervals, demonstrating the highest performance among all algorithms tested.

Conclusion

Machine learning can effectively distinguish pathological spontaneous activity in EMG audio data from voluntary activity. In particular, neural nletworks have significant potential to support clinicians in EMG diagnostics with further optimization.

Schlüsselwörter

Elektromyografie - Positive scharfe Wellen - Fibrillationen - Tiefes neuronales Netzwerk

Keywords

Positive Sharp Waves - Fibrillations - Deep Neural Network - Electromyography

Publication History

Article published online:
04 June 2025

• Literatur

• 1 Hopf HC. [Electromyography 1950-1975]. EEG EMG Z Elektroenzephalogr Elektromyogr Verwandte Geb 1976; 7: 8-13
• 2 Hopf HC, Kaeser HE, Ludin HP. et al. Basic conditions for electromyographic examination. Part I: EMG, nerve conduction velocity and test of motor end-plate function (author’s transl). EEG EMG Z Elektroenzephalogr Elektromyogr Verwandte Geb 1979; 10: 57-61
• 3 Rubin DI. Chapter 17 – Normal and abnormal spontaneous activity. In: Levin KH, Chauvel P, Hrsg. Handbook of Clinical Neurology. Elsevier; 2019: 257-279
• 4 Reig B, Heacock L, Geras KJ. et al. Machine learning in breast MRI. J Magn Reson Imaging 2020; 52: 998-1018
• 5 Mecheter I, Alic L, Abbod M. et al. MR Image-Based Attenuation Correction of Brain PET Imaging: Review of Literature on Machine Learning Approaches for Segmentation. J Digit Imaging 2020; 33: 1224-1241
• 6 Makino M, Yoshimoto R, Ono M. et al. Artificial intelligence predicts the progression of diabetic kidney disease using big data machine learning. Sci Rep 2019; 9: 11862
• 7 Huang Q, Zhang F, Li X. Machine Learning in Ultrasound Computer-Aided Diagnostic Systems: A Survey. Biomed Res Int 2018; 2018: 5137904
• 8 The Scientist & Engineer’s Guide to Digital Signal Processing, 1999 | Education | Analog Devices. Im Internet https://www.analog.com/en/education/education-library/scientist_engineers_guide.html Stand: 18.01.2019
• 9 Dinger V, Master Thesis KI. Methodiken für die Verarbeitung akustischer Signale AI Usage for Processing Acoustic Signals. 2021;
• 10 Chen CH. Signal processing handbook. New York: Dekker; 1988
• 11 Breiman L. Random Forests. Machine Learning 2001; 45: 5-32
• 12 Cover T, Hart P. Nearest neighbor pattern classification. IEEE Transactions on Information Theory 1967; 13: 21-27
• 13 Cortes C, Vapnik V. Support-vector networks. Mach Learn 1995; 20: 273-297
• 14 Natekin A, Knoll A. Gradient boosting machines, a tutorial. Front Neurorobot 2013; 7
• 15 Paass G, Hecker D. Bilderkennung mit tiefen neuronalen Netzen. 2020: 119-166
• 16 Berahmand K, Daneshfar F, Salehi ES. et al. Autoencoders and their applications in machine learning: a survey. Artif Intell Rev 2024; 57: 28
• 17 Noorbakhsh-Sabet N, Zand R, Zhang Y. et al. Artificial Intelligence Transforms the Future of Health Care. Am J Med 2019; 132: 795-801
• 18 Taha MA, Morren JA. The role of artificial intelligence in electrodiagnostic and neuromuscular medicine: Current state and future directions. Muscle & Nerve 2024; 69: 260-272
• 19 Di Nardo F, Nocera A, Cucchiarelli A. et al. Machine Learning for Detection of Muscular Activity from Surface EMG Signals. Sensors 2022; 22: 3393
• 20 Pérez-Reynoso F, Farrera-Vazquez N, Capetillo C. et al. Pattern Recognition of EMG Signals by Machine Learning for the Control of a Manipulator Robot. Sensors 2022; 22: 3424
• 21 Thamsuwan O, Johnson PW. Machine learning methods for electromyography error detection in field research: An application in full-shift field assessment of shoulder muscle activity in apple harvesting workers. Applied Ergonomics 2022; 98: 103607
• 22 Yousefi J, Hamilton-Wright A. Characterizing EMG data using machine-learning tools. Comput Biol Med 2014; 51: 1-13
• 23 Tannemaat MR, Kefalas M, Geraedts VJ. et al. Distinguishing normal, neuropathic and myopathic EMG with an automated machine learning approach. Clin Neurophysiol 2023; 146: 49-54
• 24 Nodera H, Osaki Y, Yamazaki H. et al. Deep learning for waveform identification of resting needle electromyography signals. Clin Neurophysiol 2019; 130: 617-623
• 25 Mittel D, Pröll S, Kerber F. et al. Mel Spectrogram Analysis for Punching Machine Operating State Classification with CNNs. In: 2021 26th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA). 2021: 1-4
• 26 Schulte-Mattler WJ, Georgiadis D, Zierz S. Discharge patterns of spontaneous activity and motor units on concentric needle electromyography. Muscle Nerve 2001; 24: 123-126
• 27 Robbins GT, Tucker BG, Stashuk DW. et al. Average proportional consecutive interval difference accurately differentiates spontaneous activity from motor unit potentials. Muscle Nerve 2019; 60: 566-570
• 28 Lezhenin I, Bogach N, Pyshkin E. Urban Sound Classification using Long Short-Term Memory Neural Network. 2019: 57-60
• 29 Nam S, Sohn MK, Kim HA. et al. Development of Artificial Intelligence to Support Needle Electromyography Diagnostic Analysis. Healthc Inform Res 2019; 25: 131-138
• 30 Rao KS. K E M Speech Recognition Using Articulatory and Excitation Source Features. Cham: Springer International Publishing. 2017