The goal of this work was to advance the use of muscle synergy functions and subject-general models to reduce the complexity of wearable sensor arrays and overcome the need for in-person calibration sessions by creating a more generalizable algorithm. This will contribute to the advancement of remote patient monitoring to better understand and hopefully develop techniques to prevent early onset knee osteoarthritis after anterior cruciate ligament (ACL) injury and surgery.

Abstract: Digital medicine promises to improve healthcare and enable its delivery to rural and underserved communities. A key component of digital medicine is accurate and robust remote patient monitoring. For example, remote monitoring of biomechanical measures of limb impairment during daily life could allow near real-time tracking of rehabilitation progress and personalization of rehabilitation paradigms in those recovering from orthopedic surgery. Wearable sensors have long been suggested as a means for quantifying muscle and joint loading, which can provide a direct measure of limb impairment. However, current approaches either do not provide these measures or require unwieldy wearable sensor arrays and/or in-person calibration activities that limit their use. In this thesis, I advance the use of muscle synergy functions, which leverage the synergistic relationship within a group of muscles, to reduce the complexity of wearable sensor arrays and overcome the current need for an in-person visit to a human performance laboratory for calibration. Surface electromyography (EMG) and kinematic data were recorded from leg muscles and segments of nine healthy subjects during walking. Subject-general muscle synergy models were validated using the leave-one-subject-out method for 4 different pairs of input muscle model sets using filtered EMG data. The effect of adding kinematic data (angular velocity) from thigh and shank segment locations was investigated. The average correlation between true and estimated excitations was 96% higher when angular velocity data was included in the 4-muscle input model set. The estimated excitations informed muscle activations with 6.7% mean absolute error (MAE) and 43% variance accounted for (VAF) averaged across all muscles when kinematic data was included in the model, and 7.3% MAE and 43% VAF without kinematic data. These results lay the groundwork for developing muscle synergy functions that no longer require in-person calibration, paving the way for completely remote studies of muscle and joint loading.

You can check out a recording of my defense presentation here.

This literature review was completed in support of this publication. 

Presentation given in April 2020 for Wearable Sensing course at the University of Vermont.

Background: In human biomechanics, the aim of electromyography (EMG)-driven neuromusculoskeletal (NMS) dynamics is to be able to estimate or predict muscle forces and joint moments from neural signals, or EMG measurements. Muscle activation dynamics govern the transformation from the neural signal to a measure of muscle activation. Muscle contraction dynamics characterize how muscle activations are transformed into muscle forces. The use of EMG signals is becoming more common in wearable sensor based remote patient monitoring and gait analysis. EMG driven muscle models may provide estimates of more clinically relevant biomarkers. 

Goal: The main focus of this project was to identify EMG-driven neuromusculoskeletal models describing muscle activation. Activation models were surveyed from studies that used EMG-measured excitations to drive the NMS dynamics and estimations of muscle forces and joint moments. The purpose of this project was to understand the breadth of activation models that have been used for NMS modeling, their origins, and how often they were used in literature to consider application for remote gait analysis.

Results: Activation models were surveyed from studies that used EMG measured excitations to drive the neuromusculoskeletal dynamics. Most activation models used originate with three different papers: Milner-Brown et al. (1973) The contractile properties of human motor units during voluntary isometric contractions, He et al. (1991) Feedback gains for correcting small perturbations to standing posture, and Zajac (1989) Muscle and tendon: properties, models, scaling, and application to biomechanics and motor control. The activation model from Milner-Brown et al. (1973) may be the most justified because it is supported by physiological evidence. The modeled used in He et al. (1991) was developed using previous literature. The model described in Zajac et al. (1989) was based on an unpublished analysis. A discretized version of the Milner-Brown model appears very popular in current techniques. This model was published almost fifty years ago, and I believe that with the current advancements of technology, there may be reason to revisit muscle activation models in future research.