Advancements in Electromyography (EMG) Signal Processing and Classification Techniques for Prosthetic Hand Control Applications: A Comprehensive Review | ||
| Industrial Technology Journal | ||
| Volume 3, Issue 1, 2025, Pages 59-87 PDF (1.07 M) | ||
| Document Type: Review Article | ||
| DOI: 10.21608/itj.2025.402363.1033 | ||
| Authors | ||
| Wafaa N Abdelrazik* 1; Hamed A. Ibrahim2; Ahmed El-Bialy3; Bassam A. Hemade2 | ||
| 1Electrical Department, Electronics Division, Faculty of Technology and Education, Suez University. | ||
| 2Electrical Department, Faculty of Technology and Education, Suez University. | ||
| 3Systems and Biomedical Eng, Faculty of Eng.,Cairo Univ. | ||
| Abstract | ||
| Prosthetic hand control lies at the critical intersection of biomedical engineering and rehabilitation medicine, aiming to restore natural hand function in amputees through precise interpretation of neuromuscular signals. Electromyography (EMG) has become the predominant technique for achieving intuitive control, yet despite promising outcomes in research environments, significant challenges persist in translating these advances into reliable, real-world clinical applications. To assess the current state of the field, a systematic literature review was conducted in accordance with PRISMA guidelines, drawing from PubMed, IEEE Xplore, and Google Scholar. A total of 285 studies were initially identified, of which 52 met the inclusion criteria based on classification accuracy, real-time implementation feasibility, and clinical viability. The analysis revealed that deep learning techniques consistently outperformed traditional approaches, while the integration of multimodal data and the use of advanced preprocessing methods significantly improved system robustness. However, real-time implementation introduced critical performance trade-offs, particularly in terms of latency and power efficiency. Although EMG-based control systems have reached a stage of clinical viability, especially with the superior offline performance of deep learning models, successful deployment still requires hybrid strategies that can balance high accuracy (>90%), low power consumption (<2W), and rapid response times (<300ms). Persistent barriers, such as electrode stability degradation resulting in an 18–25% drop in accuracy, and inter-session variability, underscore the necessity for adaptive calibration mechanisms to ensure consistent, long-term performance and enable widespread. | ||
| Keywords | ||
| electromyography; signal processing; machine learning; deep learning; gesture recognition | ||
|
Statistics Article View: 79 PDF Download: 16 |
||
