Nanomembrane Wireless Graphene Biopatch with Machine Learning for Detection of Muscle Activities and Determination of Intended Motions

Prediction of the human intended movement based on a non-invasive manner is of great significance due to its direct and wide-range applicability to human-machine interfaces and interactions. However, predicting the intended motion involves engineering challenges because the dynamic movements and highly elastic nature of human skin surface usually induce motion-related noise when recording electromyography (EMG) signals with conventional rigid sensors. Here, we introduce a nanomembrane, wireless EMG sensor system based on screen-printed graphene and a thin-film active circuit to predict the intended human upper extremity movement in real-time by incorporating the embedded deep learning algorithm.

Owing to the nanomaterials and stretchable mechanics design, the integrated set of electrodes forms a fully conformal contact with human skin's irregular and elastic surface, thereby minimizing motion artifacts while facilitating continuous detection of EMG signals. In addition, the thin-film, flexible circuit with Bluetooth offers a high-fidelity recording of signals while avoiding wire-induced signal artifact. Moreover, the embedded IMU (inertial measurement unit) in the biopatch that can monitor linear acceleration and the upper body part orientation in real-time provides the trainable data for deep learning, allowing more accurate determination of the intended motions. The systematic detection of multi-EMG and IMU signals from different muscle parts of the human body collects the trainable dataset for convolutional neural networks based deep-learning algorithm, which accurately determines a user’s intention with over 95% accuracy of 6 different classes.