Fuzzy MUAP recognition in HSR-EMG detection basing on morphological features

• Published in: Journal of electromyography and kinesiology Vol. 24; no. 4; pp. 473 - 487
• Main Authors: Mebarkia, Kamel, Bekka, Raїs El’hadi, Reffad, Aicha, Disselhorst-Klug, Catherine
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01.08.2014
• Subjects: Action Potentials - physiology; Adult; Algorithms; Electromyography - methods; Female; Fuzzy classification; Fuzzy Logic; Fuzzy ‘if-then’ rules; Hand; High spatial resolution EMG; Humans; Laplacian filter; Male; Motor unit action potential; Muscle, Skeletal - physiology; Optimization; Parameterization; Pattern recognition; Physical Medicine and Rehabilitation; Probability; Signal Processing, Computer-Assisted; Young Adult; Fuzzy ‘if-then’ rules; High spatial resolution EMG; Laplacian filter; Motor unit action potential; Pattern recognition; Parameterization; Fuzzy classification; Optimization
• ISSN: 1050-6411; 1873-5711; 1873-5711
• DOI: 10.1016/j.jelekin.2014.04.006

The idea of ‘besides the MU properties and depending on the recording techniques, MUAPs can have unique pattern’ was adopted. The aim of this work was to recognise whether a Laplacian-detected MUAP is isolated or overlapped basing on novel morphological features using fuzzy classifier. Training data set was constructed to elaborate and test the ‘if-then’ fuzzy rules using signals provided by three muscles: the abductor pollicis brevis (APB), the first dorsal interosseous (FDI) and the biceps brachii (BB) muscles of 11 healthy subjects. The proposed fuzzy classier recognized automatically the isolated MUAPs with a performance of 95.03% which was improved to 97.8% by adjusting the certainty grades of rules using genetic algorithms (GA). Synthetic signals were used as reference to further evaluate the performance of the elaborated classifier. The recognition of the isolated MUAPs depends largely on noise level and is acceptable down to the signal to noise ratio of 20 dB with a detection probability of 0.96. The recognition of overlapped MUAPs depends slightly on the noise level with a detection probability of about 0.8. The corresponding misrecognition is caused principally by the synchronisation and the small overlapping degree.