Suitability of Single-channel Acoustic Myography for Classification of Individual Finger Movements

Acoustic Myogram (AMG) is the vibration or sound signal produced during muscle contraction and relaxation. A simple system like a condenser microphone is enough to capture an AMG signal from muscles, unlike complex systems that are used in surface Electromyography (sEMG). Moreover, AMG signal is not highly sensitive to sensor placement like sEMG signal. Therefore, Acoustic Myography is a potential research area to find an alternative to complex and bulky sEMG system. This work focuses on verifying the suitability of single-channel AMG for the classification of individual five finger movements. AMG data were recorded from 14 subjects at two different sites on hand, namely forearm muscle and wrist. Temporal, Spectral, and Cepstral features were extracted from the collected data after required pre-processing. Two Machine learning algorithms: Support Vector Machine and K-Nearest Neighbors were applied to classify the features. From this analysis, three main outcomes were achieved: Independent five finger movements cannot be differentiated precisely using single-channel AMG data solely, whether they are from the forearm or wrist. Class reduction and grouping of some fingers increase the classification accuracy, which infers that vibrations due to different finger movements have very similar attributes. AMG from both forearm and wrist yielded similar classification accuracy, with no evidence of a site being significantly better. Contribution: This research reveals the competence of single-channel Acoustic Myography (AMG) in classifying individual hand finger movements.