Wearable Neuromotor Sensing

Abstract

Movement is essential to human function and life. Wearable neuromotor sensing enables accessible and non-invasive study of the systems that govern human movement. In this thesis, I present three approaches in materials science, in electronics, and in multimodal sensor fusion for biomechanics, to enhance wearable neuromotor sensing.

This thesis begins by improving signal-to-noise of surface electromyography (EMG). I develop a novel preclinical ex-vivo model to experimentally isolate the bioelectrochemical features of single skin-electrode contact. In this model, soft conductive polymer hydrogels made of PEDOT: PSS present nearly an order of magnitude decrease in the skin- electrode contact impedance (88%, 82%, and 77% at 10Hz, 100Hz, and 1kHz, respectively) when compared to clinical electrodes. Integrating these soft conductive polymer blocks into an adhesive wearable sensor increases the EMG signal-to-noise ratio (average 2.1dB increase, max 3.4dB increase) when compared to clinical electrodes across all human subjects. I demonstrate the utility of this higher fidelity EMG in a neural interface system: EMG-based velocity-control of a robotic arm to complete a pick and place task.

This thesis then expands what can be sensed with EMG electrodes. Changes in electrode-skin conditions due to contact pressure variation, sweat, and dehydration lead to variation in bioimpedance across skin locations and thus variation in the EMG voltages measured at skin electrodes. By combining analog circuits, digital signal processing, and analytic calculations using bioimpedance principles, a novel system enables the decoupling of EMG and bioimpedance signals by simultaneously measuring both signals with the same electrodes already used for EMG. Design rationales for the system are explicitly defined and benchtop characterizations show accurate bioimpedance measurements (R^2 ~ 0.96) under carefully controlled EMG-like signals from a function generator. I demonstrate system utility in vivo during controlled force generation tasks where controlled alteration to subjects’ skin-electrode conditions produce changes in both EMG and bioimpedance. Finally, this thesis leverages multimodal sensor fusion machine learning to fuse EMG and muscle ultrasound imaging for a critical movement application: balance. Elderly non-fatal falls from balance loss cost American society $50 billion in direct healthcare costs. Ultrasound enables muscle state tracking, especially of deep musculature not accessible to surface EMG. I design a novel multi-stage machine learning architecture with kinematics, ultrasound, and EMG sensing to forecast and to estimate the ankle torques subjects generate in single leg balance when perturbed anteriorly, medially, and laterally. The pipeline results in 6% normalized RMSE in both the sagittal and the frontal plane torque estimation. Furthermore, the pipeline forecasts torques 74 milliseconds in the future even under the influence of perturbations, opening new avenues for balance assistance and diagnosis where future human intent may be useful or essential.

I present preliminary work for a stretchable adhesive EMG array embedded with a flexible ultrasound probe. Such a device would benefit from the bioelectronic performance of PEDOT:PSS. Data from the device could be directly leveraged by the multimodal machine learning pipeline for balance control or other applications in human intent.

Ultimately, my thesis aims to build accessible tools for researchers interested in neuromotor sensing. I build from basic preclinical characterization to applied machine learning forecasting on human subject data. I hope this thesis helps future researchers in their works and encourages readers to mix fields to improve their science.