Advances in technology and research have been employed in recent years to develop efficient mechanisms to deliver home-based exercise therapy to patients suffering from knee osteoarthritis, a degenerative disease associated with aging. Essential to the success of a therapeutic home-exercise program is the quality of the motion performed by the patient. The unsupervised nature of home-based exercise may lead to incorrect exercise performance by patients; however, current home-based exercise programs do not provide mechanisms for monitoring the quality of motion performed or for providing feedback to the patient. This lack of support has been found to be a factor in patient non-compliance to home exercise programs. Our goal is to provide a motion sensor-based system that can evaluate the quality of exercise to support home rehabilitation. We introduce the Quality Assessment Framework (QAF) that uses low-cost motion sensors with data processing and machine learning techniques to assess the quality of human motion performed during therapeutic exercises. Data from fifteen persons with knee osteoarthritis were collected in a laboratory environment, and a classifier was trained using multi-label learning methods to detect descriptive characteristics of the patient's motion. These characteristics represent errors in the exercise performance as well as variables, such as speed, that are regularly monitored by the patient's therapist. Results from multi-label learning are presented and recommendations are made on requirements for an in-home therapeutic exercise system. A classifier, using Ensembles of Classifier Chains with a Support Vector Machine base classifier, provides the best method for assessing human motion quality in the QAF. Leave-one-out and leave-half-out testing provided us with information on the achievable level of generalizability for new patients whose motion is not contained in the training set. We found that a small amount of new patient data is required for good recognition of characteristics in exercise performances. The QAF can be adapted to the home therapy needs of conditions other than knee OA. We present a preliminary design of the InForm Exercise System that utilizes the QAF and has the potential to present feedback to patients completing home exercise programs.

Calculating femoral bone density changes after hip arthroplasty is of interest to researchers and clinicians for predicting the longevity of the prosthetic implant and the surrounding bone. Recently clinicians have been administering bisphosphonate drugs in an attempt to reduce the bone resorption due to stress shielding caused by these implants. Current strain-adaptive computational models with bisphosphonate treatment don’t predict the long term effects or look at treatment with hip resurfacing implants. The main goal of this study was to create and validate a computer model of the human femur incorporating a bone remodeling algorithm based on biological remodeling processes and bisphosphonate drug treatment. A secondary objective was to then create various bisphosphonate drug treatment scenarios and evaluate differences in bone density, damage, and activation frequency. Experimental studies were used to validate the model and the effects of bisphosphonates. A finite element model created from a CT scan of a cadaveric femur, a bone remodeling algorithm, and a bisphosphonate algorithm were incorporated into the model with loading conditions representative of walking and stair climbing. The model was allowed to evolve from an initial state of homogenous density to a steady state form with a density similar to that of the femur. Reduced loading representative of decreased muscle forces were applied to the steady state form to simulate preoperative conditions of a patient with hip osteoarthritis. Both a femoral hip resurfacing component and an uncemented, tapered stem were then integrated in the computer model representative of a postoperative state. Bisphosphonate treatment was applied to both the preoperative and postoperative states in several scenarios after untreated simulations. Bone loss was predicted over a six year postoperative period for both implants and varying treatments. Femoral bone loss in bisphosphonate treatment scenarios matched results seen clinically. Bone volume fraction (BVF) showed little change between one year preoperative to one year postoperative Alendronate treatment and one year postoperative Alendronate treatment for a specific implant type. Both treatment scenarios increase the BVF over no treatment. Pretreating with Alendronate appears to help against femoral neck fracture. This study successfully created a three-dimensional finite element model able to simulate long term effects of the remodeling process in bone with Alendronate treatment. The results show an importance of treatment timing for both types of implants especially when potentially requiring a revision surgery.

This study investigated reducing volumetric flowrate under steady flow conditions by varying lengths of compression with constant cross-sectional area and varying cross-sectional area reduction with constant length in order to better understand how to control junctional hemorrhaging. The hypotheses of this study were that length reduction will have little effect on volumetric flowrate and that cross-sectional area reduction would need to be approximately 80 percent in order to obtain bleeding control. The study found that length reduction has little effect on changing the flowrate. However, in order to obtain at least 80 percent reduction in flow, the area needs to be occluded by at least 95 percent. These results may help inform better tourniquet designs by using collapsible tube science.

Osteoporosis affects nearly 54 million people in the United States. The cost associated with treatment is estimated to be $19 billion per year and is expected to grow yearly. D-spacing is the staggering of collagen molecules found at the nanoscopic level. Previously thought to have a constant value, recent studies have found that D-spacing has a distribution of values throughout the tissue. As part of an ongoing effort in understanding the mechanisms that are affected by osteoporosis, a finite element model was developed to explore the effects of D-spacing distribution on the viscoelastic material properties of bone tissue. The goal of this computational model was to mimic the viscoelastic properties of different sectors of bone tissue that have been treated under different loading conditions (tension and compression). An appropriate animal model was required to allow for the development of an accurate computational model. Although they don't exhibit similar hormonal cycles as humans, sheep are an excellent animal model for bone research as they experience Haversian bone remodeling, are docile, relatively inexpensive, and have skeletons similar in size and mechanical properties to humans. For this study, six Rambouillet-cross ewes were either ovariectomized (OVX) or underwent a sham surgery (control). After twelve months post-surgery, the ewes were euthanized and rectangular beam bone samples were collected from different sectors of the ulna/radius bones. Dynamic mechanical analysis was performed on these samples and the viscoelastic property, tangent delta, was measured from each analysis at varying frequencies. Using experimental measurements, the Composite Model was developed on finite element analysis software, Abaqus. The model was generated through a Python script that uses experimental D-spacing mean and standard deviation data to create a large two-dimensional model composed of two hundred collagen and hydroxyapatite complexes with varying D-spacing lengths. Multiple security measurements were implemented to ensure biological relevance. Collagen was assigned viscoelastic material properties through a user subroutine material property. Four models for each sector of interest (caudal and cranial) were generated. Each model was loaded under appropriate loading conditions and tangent delta was recorded for each test frequency. Results from the Composite Model matched the experimental data more accurately than previous computational models, suggesting a superior model. The results implied that a large network of collagen and hydroxyapatite complexes in series and parallel are effective at modeling bone under different loading conditions. This computational model shows promise in the bone research field. A lot of flexibility was implemented in the model development process, making refinements easy to be performed. This study provides a stepping-stone in computational tooling on examining the effects of metabolic bone diseases on viscoelasticity.

<p>A dynamic ballistic elbow extension task was chosen to investigate the mechanisms responsible for achieving a high final velocity during this type of movement. After a screening process of one hundred male participants, thirty-two were chosen to be further investigated who fell into the extremes of the strength and speed continuums. The main investigation involved having participants ballistically extend their elbows against external relative loads of 0%, 20%, 40%, 60% and 80% of their maximum isometric force and two absolute loads of 1.1 kg and 2.2 kg. External torque and angular displacement measurements were recorded as well as triceps and biceps electrical activity. EMG modeling, which employed the characteristics of muscle mechanics, was used to determine the differences in performance. Isometric strength did not produce a 1:1 mapping with maximum velocity. Individuals existed that were relatively strong but not fast. Additionally, there were subjects that were quite strong but not appreciably fast. Peak instantaneous power, however, produced the best correlations with peak final velocity. To determine why certain individuals were capable of producing more power, each subject's triceps EMG was modeled in order to predict the actual muscle torque. The model predicted torques with a mean correlation of 0.957 and a mean RMSerror value of 5.8 Nm for the 224 trials. As a result of the good predictions, a forward-dynamics approach was used to manipulate weaknesses in one performance with another individual's superior attributes. Performance improvements were noted as a result. These findings demonstrate that peak instantaneous power best predicts peak terminal velocity. Furthermore, this study has developed a model capable of identifying neuro-muscular weaknesses in performance and suggesting how improvements in those areas would change the maximum velocity attained. The next stage is to determine the proper training stimuli that would make these specific neuro-muscular improvements possible.</p> / Doctor of Philosophy (PhD)

Mobile brain-body imaging (MoBI) seeks to understand human brain and body dynamics during movement and locomotor tasks such as walking with perturbations that challenge balance and lead to adaptation of walking behavior. In this dissertation, I evaluated the long-term electromyography (EMG) recording performance of dry epidermal electrodes for measuring electrical muscle activity. I also evaluated the relationships between the signals recorded from the two sides of dual-sided electroencephalography (EEG) electrodes, a recent advancement in EEG electrode design for measuring electrical brain activity. Last, I investigated adaptation of brain and body responses to small and frequent perturbations during treadmill walking while I recorded brain activity using a custom-built dual-layer EEG system and body kinematics using motion capture. Dry epidermal electrodes provided better Signal Quality Indices, a metric I developed that accounts for signal-to-noise and signal-to-motion contributions, during limited dynamic movements, indicating that high-quality EMG for long-term recording was possible but also limited. For the dual-sided EEG electrode evaluation, I quantified correlations between dual-sided EEG signals in a benchtop experiment. Signals recorded from two sides of a dual-sided EEG electrode were highly correlated during constrained movements but degraded in more realistic random movements. This information is critical for developing EEG cleaning algorithms based on dual-layer EEG systems. For the locomotor adaptation studies, I quantified gait stability using margin of stability and its components and performed source localization and time-frequency analyses to determine electrocortical processes during perturbed walking. Small and frequent treadmill perturbations disrupted gait stability and quickly induced direction-dependent gait stability adaptation. Anterior cingulate theta-band adaptation occurred and was more evident during belt deceleration perturbations compared to belt acceleration perturbations. These results add new knowledge about the characteristics of novel EMG and EEG electrodes and revealed the potential of modulating perturbation direction to tune gait stability strategy and activation of electrocortical dynamics.

On a daily basis, humans walk over a variety of terrains. Studies have shown that spatiotemporal gait parameters, such as stride length, stride frequency, stride variability, etc., change when humans walk down a decline and up an incline compared to level ground. However, these studies have been limited to using fixed speed treadmills or analyzing a small number of strides when conducted over ground. Thus, there is a need to investigate the fluctuations in spatiotemporal gait parameters of humans walking at their self-selected speed, which requires recording hundreds of strides. Here we hypothesized that subjects will walk with a slower speed and have greater stride variability on an incline or decline compared to level ground. We used a self-paced treadmill and had 7 young adults walk on three slopes (+9 degrees, incline; 0 degrees, level; -9 degrees, decline). A motion capture system was used to calculate spatiotemporal gait parameters. The results showed that subjects walked the fastest on level ground (1.15 +/- 0.17 m/s). Subjects walked more slowly during decline walking (1.06 +/- 0.14 m/s) and walked the slowest during incline walking (0.92 +/- 0.18 m/s). There was not a single steady-state speed that subjects used for all slopes. Instead, there were multiple periods when the subject was not at a steady state. Only ~60% of the strides could be classified as being at steady-state. When walking down a decline, subjects needed ~10 +/- 1 more strides to reach the first steady-state period. When walking on an incline and decline, stride length variability increased by ~1.6x (0.0014�2 ± 0.0008�2) and ~1.2x (0.0012�2 ± 0.0008�2 ) compared to level ground (0.0005 �2 ± 0.0003 �2). Stride width variability increased by ~20.6x (0.0108�2 ± 0.0121�2 ) and ~14.2x (0.0076�2 ± 0.0044�2 ) for incline and decline slopes compared to level ground (0.0005 �2 ± 0.0003 �2). These results provide greater insight on the fluctuations during self-selected walking speeds subjects use on different slopes. This could have implications on balance control and fall risk during walking.

This work is an experimental study and fabrication of design concepts to validate the feasibility of smart materials and their applications in bio-inspired robotics. Shape-Memory Alloy (SMA) springs are selected as the smart material actuators of interest to achieve locomotion in the proposed mobile robot. Based on a previous design of the robot, this work focuses on both implementing a new locomotion concept and reducing size and weight of the previous design, both using SMA based actuators. Objectives are met in consideration of the conceptual mechanics of circular robot locomotion. The first prototype is a variation of the original design. It consists of a soft, rubber outer shell with three intrinsically attached diametric SMA springs that deform the outer shell during contraction and relaxation. The springs were provided with electrical current in patterns to produce deformation needed to generate momentum and allow the robot to tumble and roll. This design was further improved to provide more stability while rolling. The second design concept is a modification of our previous design leading to reduction in size and weight while maintaining essentially the same mechanism of locomotion. In this case, the SMA springs were externally configured between the end of equi-spaced spokes and the circular core. Upon actuation, the spokes function as diametrically translating legs to generate locomotion. These design concepts are fabricated and experimented on, to determine their feasibility, i.e. whether rolling/tumbling motion is achieved. The scope of the project was limited to demonstration of basic locomotion, which was successful. Future work on this project will address the design of automatic control to generate motion using closed-loop sensor-based actuation.

Metabolic power and cost of transport (COT) are common quantifiers for effort when performing tasks including walking and running. Most studies focus on using a range of normal walking speeds over level ground or varied slopes. However, these studies use fixed-speed conditions. Fatigue, stability, metabolic expenditure, heart rate, and many other factors contribute to normal walking speed varying over time. This study aimed to show that allowing a subject to walk with a self-paced speed should correlate to a minimum COT at a given slope. This study also aimed to determine if a preferred slope exists based on minimizing metabolic expenditure or maximizing stability. In this study, subjects walked at four different speed conditions including three fixed speeds (0.75 m/s, 1.0 m/s, 1.25 m/s) and their self-paced speed at five different slopes (-6°, -3°, 0°, 3°, 6°) while metabolic energy expenditure and motion were recorded. The minimum COT occurred at a 3° decline. At this slope, some subjects preferred to walk at a faster speed compared to level ground, whereas other subjects walked with a slower speed compared to level ground. Thus, there was a greater range of self-paced speeds, from 0.745 m/s-2.045 m/s. In comparison, at a 6° incline, the range of self-paced speeds was much smaller, from 0.767 m/s-1.434 m/s. The variance among self-paced speeds and slope conditions between subjects suggests that COT, alone, does not explain walking decisions; stability might play a greater role than initially believed. These results provide greater insight into why humans choose to walk at a certain speed over a range of slopes and terrains.

Electroencephalography (EEG) non-invasively records electrocortical activity and can be used to understand how the brain functions to control movements and walking. Studies have shown that electrocortical dynamics are coupled with the gait cycle and change when walking at different speeds. Thus, EEG signals likely contain information regarding walking speed that could potentially be used to predict walking speed using just EEG signals recorded during walking. The purpose of this study was to determine whether walking speed could be predicted from EEG recorded as subjects walked on a treadmill with a range of speeds (0.5 m/s, 0.75 m/s, 1.0 m/s, 1.25 m/s, and self-paced). We first applied spatial Independent Component Analysis (sICA) to reduce temporal dimensionality and then used current popular classification methods: Bagging, Boosting, Random Forest, Naïve Bayes, Logistic Regression, and Support Vector Machines with a linear and radial basis function kernel. We evaluated the precision, sensitivity, and specificity of each classifier. Logistic regression had the highest overall performance (76.6 +/- 13.9%), and had the highest precision (86.3 +/- 11.7%) and sensitivity (88.7 +/- 8.7%). The Support Vector Machine with a radial basis function kernel had the highest specificity (60.7 +/- 39.1%). These overall performance values are relatively good since the EEG data had only been high-pass filtered with a 1 Hz cutoff frequency and no extensive cleaning methods were performed. All of the classifiers had an overall performance of at least 68% except for the Support Vector Machine with a linear kernel, which had an overall performance of 55.4%. These results suggest that applying spatial Independent Component Analysis to reduce temporal dimensionality of EEG signals does not significantly impair the classification of walking speed using EEG and that walking speeds can be predicted from EEG data.