Measurement and estimation of muscle contraction strength using mechanomyography based on artificial neural network algorithm

Muscle contraction strength estimation using mechanomyographic (MMG) signal is typically calculated by the root mean square (RMS) amplitude. Raw MMG signal is processed by rectification, low-pass filtering, and mapping. In this work, beside RMS amplitude, another significant parameter of MMG signal, i.e. frequency variance (VAR), is introduced and used for constructing an algorithm for estimating the muscle contraction strength. Seven participants produced isometric contractions about the elbow while MMG signal and generated torque (resultant of muscle contraction strength) of biceps brachii were recorded. We found that MMG RMS increased monotonously and VAR decreased under incremental voluntary contractions. Based on these results, a two-layer neural network was utilized for the model of estimating the muscle contraction strength from MMG RMS and VAR. Experimental evaluation was performed under constant posture and sinusoidal contractions at 0.5 Hz, 0.25 Hz, 0.125 Hz, and random frequency. The results of the proposed algorithm and MMG RMS linear mapping were also compared. The proposed algorithm has better accuracy than linear mapping for all contraction frequencies. The mean absolute error decreased 6% for the 0.5Hz contraction, 43% for 0.25 Hz contraction, 52% for 0.125 Hz contraction, and 30% for random frequency contraction.