Common spatial-spectral analysis of EMG signals for multiday and multiuser myoelectric interface. (August 2019)

• Record Type: Journal Article
• Authors: Sheng, Xinjun; Lv, Bo; Guo, Weichao; Zhu, Xiangyang
• Is Part Of: Biomedical signal processing and control. Volume 53(2019)
• Publication Date: 2019-08
• Subjects: Myoelectric interface -- Common spatial-spectral analysis -- Multiday zero retraining -- Multiuser zero training; 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.2019.101572
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A novel signal processing method named common spatial-spectral analysis (CSSA) is applied to EMG signals. CSSA is used to find common modes across multiday or multiuser EMG signals. The proposed preprocessing technique (CSSA) improves the classification performance on multiday and multiuser scenarios. CSSA is recommended for multiday zero retraining or multiuser zero training myoelectric interface. Abstract: Practical implementation of myoelectric interfaces have been largely hindered by cumbersome training and retraining procedures required for use across multiple days or for multiple users. We thus present a common spatial-spectral analysis (CSSA) framework to eliminate the need for retraining over multiple days or for multiple users. The CSSA is implemented through spectral decomposition and common modes analysis, maximumly utilizing the common spatial-spectral electromyography (EMG) mode from multiple days or for multiple users. Experiments involving two scenarios were conducted to simulate the application of multiday or multiuser myoelectric interface. Eight healthy and three amputee subjects participated in the first experiment for ten consecutive days, and seven healthy subjects participated in the second experiment involving a multiuser interface. Experimental results demonstrated that the control performance without retraining the myoelectric interface with the CSSA was significantly improved, and the classifier model pre-trained by background data under CSSA enabled EMG signals from new days or users to be recognized without training or retraining. The results could serve as a foundation for practical implementation of myoelectric interfaces.