4th Annual Mountain West
Biomedical Engineering Conference

The effective and intuitive control of a neural prosthetic hand requires access to a sufficient number of stable and reliable control signals. Implantable myoelectric sensors (IMES), which can record independent EMG signals wirelessly without percutaneuos connectors, could be used to obtain such control information. Decode algorithms can be applied to these EMG signals to determine the intended movement and drive a prosthetic limb. IMES were implanted into the hand muscles in the forearm of a rhesus monkey to test the efficacy of IMES for this application. EMG signals were recorded from the IMES while the monkey performed random individuated and combined finger flexions of the thumb, index, and middle fingers on a manipulandum. A parallel linear discriminant analysis (LDA) based algorithm was implemented to determine if there was a movement and, if so, which finger was moved. The decode implemented multiple classifiers, thus allowing multiple classes to be active simultaneously. These algorithms performed an LDA on extracted time domain features of the EMG such as zero crossings, slope changes, waveform lengths, and mean absolute values from the continuously windowed EMG signals. Both algorithms were trained on four finger states (no motion, thumb, middle, and index finger flexion) based on the switch closure data sensed by the manipulandum. The algorithms then continuously predicted the finger states of a testing data set without using the switch closure data. These LDA based algorithms were accurate in correctly decoding when and which individual finger was pressed. These results demonstrate that simple algorithms can use wireless EMG data to effectively infer individual finger flexions for dexterous control of an artificial hand.