Compliant Magnetic Sensor Arrays Enable Real-Time Force Myogram Pattern Recognition for Dexterous Hand Control by Amputees

Control of powered prosthetic hands could benefit from improved techniques to infer the amputees' desired grasp intentions. Force myography (FMG) has recently emerged as a potential alternative to electromyography, which is traditionally used in clinical practice. Thus, we introduce innovative compliant magnetic FMG sensor arrays for forearm muscle pattern recognition that enables ten subjects, including three upper limb amputees, to have real-time control of a dexterous artificial hand. Sensor arrays with 18 or 24 compliant magnetic FMG sensor modules are fabricated, demonstrating the customizability of our 3D scanning-3D printing process to create individualized wearable sensor arrays for varying limb differences. On average, all 10 subjects control $16.20~\pm ~3.68$ classes with mean accuracy of 93.64% $\pm ~2.90$ % in real-time control experiments. Additionally, principal component analysis (PCA) is used to both select the most impactful sensors and improve classification accuracy in subsequent offline analyses. Seven out of the ten subjects achieve their highest accuracy with a reduced number of compliant magnetic sensors, highlighting the value of the PCA-informed sensor selection approach. Because the proposed compliant magnetic sensor arrays are highly practical, can function underwater, and have a signal to noise ratio (SNR) of 34.51 dB $\pm ~1.49$ dB, we are providing our open-source dataset of FMG signals from 3 amputees and 7 non-amputees as a resource for the research community. These novel compliant magnetic FMG sensor arrays have the potential to advance the state of the art for prosthetic hand control and could be used broadly in the fields of tactile sensing, haptics, medical robotics, and rehabilitation.