Steven R. Anton

Smart knee replacement technology seeks to provide an in vivo method of tracking long-term postoperative joint loads with the goal of identifying clinically relevant phenomena linked to postoperative dissatisfaction in real time. This study evaluated the ability of a piezoelectric compartmental force and compartmental center of pressure sensing total knee replacement to sense condylar lift-off, which is a clinically relevant phenomenon commonly attributed to postoperative dissatisfaction. A commercially available total knee replacement was modified to include six piezoelectric transducers capable of measuring compartmental forces and tibiofemoral centers of pressure on the articular surface of the tibial bearing insert. The smart knee replacement was evaluated with a six-degree-of-freedom joint motion simulator applying a varus lift-off profile. The study demonstrated that the lift-off was evident in both the sensed joint loads and the localized tibiofemoral centers of pressure obtained from the piezoelectric sensing system. The results indicated that the piezoelectric smart knee replacement could be effective for detecting this clinically problematic mechanical issue.

Machine learning (ML) is capable of aiding and improving medical diagnostics for a wide variety of pathologies. When used to process data from smart implantable medical devices, ML can offer timely and automated diagnostic tools to improve patient care. However, diagnostic ML models that influence healthcare decisions must have outputs that can be understood and trusted. Currently, ML-based medical diagnostic research focuses mainly on accuracy, lacking investigation of interpretability, explainability, and trust in the models — fundamental principles of diagnostic ML. To address this gap, this study seeks to improve explainability in ML models trained to diagnose aseptic tibial loosening in smart piezoelectric total knee replacements (TKRs). Specifically, a pathway to explainability is presented by applying the local interpretable model-agnostic explanations (LIME) method to interpret previously trained [ ]-nearest neighbor (KNN), support vector machine (SVM), and discriminant analysis (DA) ML models that classify cement damage from electromechanical impedance (EMI) signatures of piezoelectric-instrumented simulated TKRs. Two simple yet novel feature engineering techniques are proposed to align the models with domain knowledge of piezoelectric impedance-based structural health monitoring (SHM), thus improving explainability. These feature engineering techniques are broadly applicable to data types where adjacent features have inherent relations (i.e., time series, spectra, images, etc.). The original KNN, SVM, and DA models demonstrated explainability scores of 52%, 64.4%, and 42.1%, respectively. The first feature engineering technique improved the KNN and SVM scores to 84.6% and 87.6%, respectively, with DA falling to 29.5%. The second feature engineering technique improved the KNN and SVM scores to 74.5% and 81.1%, respectively, with DA falling to 33.5%. The pathway to explainability and the feature engineering techniques presented in this study yield models with improved explainability that are better aligned with domain knowledge, while either improving (SVM) or maintaining (KNN) their original accuracy.

In orthopedics, the evaluation of footbed pressure distribution maps is a valuable gait analysis technique that aids physicians in diagnosing musculoskeletal and gait disorders. Recently, the use of pressure-sensing insoles to collect pressure distributions has become more popular due to the passive collection of natural gait data during daily activities and the reduction in physical strain experienced by patients. However, current pressure-sensing insoles face the limitations of low customizability and high cost. Previous works have shown the ability to construct customizable pressure-sensing insoles with capacitive sensors using fused-deposition modeling (FDM) 3D printing. This work explores the feasibility of low-cost fully and continuously 3D printed pressure sensors for pressure-sensing insoles using three sensor designs, which use flexible thermoplastic polyurethane (TPU) as the dielectric layer and either conductive TPU or conductive polylactic acid (PLA) for the conductive plates. The sensors are paired with a commercial capacitance-to-voltage converter board to form the sensing system. Dynamic sensor performance is evaluated via sinusoidal compressive tests at frequencies of 1, 3, 5, and 7 Hz, with pressure levels varying from 14.33 to 23.88, 33.43, 52.54, and 71.65 N/cm² at each frequency. Five sensors of each type are tested. Results show that all sensors display significant hysteresis and nonlinearity. The PLA-TPU sensor with 10% infill is the best-performing sensor with the highest average sensitivity and lowest average hysteresis and linearity errors. The range of average sensitivities, hysteresis, and linearity errors across the entire span of tested pressures and frequencies for the PLA-TPU sensor with 10% infill is 11.61–20.11·10⁻⁴ V/(N/cm²), 11.9–31.8%, and 9.0–22.3%, respectively. The significant hysteresis and linearity error are due to the viscoelastic properties of TPU, and some additional nonlinear effects may be due to buckling of the infill walls of the dielectric.

In this study, a new algorithm is proposed to improve impact event localization for vibration-based smart building applications. Vibration-based smart building systems, which use floor-mounted accelerometers, have the potential to be a less intrusive way to monitor building occupants. Vibration-based localization algorithms typically use a time or energy-based approach or a combination of the two. Time-based approaches rely on time-difference-of-arrival (TDOA) calculations and can have large errors due to the variability of the wave velocity through the floor. The variability in wave velocity is caused by the dispersive properties of concrete, wave reflections from walls, and discontinuities in the floor structure such as beams under the floor. Energy-based localization methods have been shown to overcome the problem of variable wave velocity by using an exponential decay model of the generated wave energy to predict the location of an impact. Previous energy-based methods have been shown to give sub-meter accuracy for measurement systems with accelerometers mounted on the surface of the floor as well as systems with accelerometers mounted under the floor. However, available energy-based methods are limited to exponential decay models, which are only applicable to homogeneous floor structures. Supporting floor structures, such as beams, and walls create inhomogeneities, which cause the wave to decay differently in each direction and require a more complex model to describe. This work proposes a new velocity and energy ratio mapping (VERM) localization algorithm to overcome this limitation. The VERM algorithm computes ratios of response energy between pairs of sensors for impact locations across the floor structure to be monitored and uses the resulting energy ratio maps to find potential impact locations. The VERM algorithm predicts a final impact location using velocity maps to compare estimated TDOA values to the TDOA value of the measured signal. Experimental validation of the method is performed by collecting hammer impact data from 70 locations in a classroom within the Lab Science Commons Building at Tennessee Technological University from three under-floor-mounted sensors and creating the energy ratio maps. Results of this study show that the VERM algorithm has an accuracy, average distance error, and maximum distance error of 77.14%, 1.89 ft, and 12.34 ft, respectively. Additionally, the results show that the algorithm produces a set of potential locations where at least one potential location is correct 90% of the time showing the algorithm can be improved to have higher accuracy.

The ability of traditional tensile tests to determine mechanical properties of human tissue can be affected by several factors, which may include failure within the clamping area of standard clamp designs. The use of custom-designed 3D printed clamps seems to provide success in preventing tissue failure and minimizing slippage during testing. Additionally, due to the nonhomogeneous characteristics of tissue, the mechanical properties often vary along the length of specimens, therefore a full-field measurement technique is required to capture the complete behavior of the tissue. Non-contact imaging techniques can be used to quantify full-field deformation during the test, allowing behavior to be quantified in different regions of the sample. Digital image correlation (DIC) is investigated in this work and is able to calculate strain by comparing successive images taken during the sample deformation. In order to measure changes in the pictures, the analyzed surface must have a well-defined high-contrast random pattern. Currently, methods for tensile testing of human tissues combined with DIC typically use an airbrush to create a random pattern on the tissue surface. Replication of this existing technique can be challenging based on the lack of control of the paint applied by the airbrush. This work presents an approach for tensile testing of well-preserved human tissue using 3D printed clamps and DIC. The samples investigated in this work include plantar fascia and Achilles tendons to analyze the performance of the proposed clamp design in thin and thick human tissue. The failure of thin and thick well-preserved human tissue was reached with a maximum load of 1124.9 N and 2167.08 N, respectively. The 3D printed clamps showed signs of clamping-related failure during the test even after increasing the infill percentage and adjusting the clamp layer orientation. The samples were dyed in methylene blue and painted with a user-defined random pattern. A custom resin roller generated by OpenSCAD was used to paint the surface of the plantar fascia allowing standardization of the creation of a random pattern on the sample surface. However, the curvature of the Achilles tendon in the most distal region where it attaches to the calcaneus bone (i.e. heel bone) precluded use of the roller. Therefore, a random pattern was hand-painted instead. The hand-painted random pattern had a performance similar to the pattern made by the resin roller which followed commercial roller characteristics. Results of this study can be used in the future to help standardize tensile testing techniques for mechanical evaluation of human tissue.

Total knee arthroplasty (TKA) is currently one of the most common orthopedic surgeries worldwide. While TKA is generally successful, revision due to pain and failure is inevitable and adversely impacts patient outcomes. Catastrophic failure rates may be reduced by early detection of damage. This work evaluates the ability of machine learning to detect and classify damage from piezoelectric impedance measurements of total knee replacements (TKRs). Multiple simulated TKR test samples are constructed and artificially damaged to varying degrees, and impedance spectra are recorded. Five machine learning architectures are examined including decision trees, k-nearest neighbor (KNN) classifiers, discriminant analysis classifiers, naïve Bayes classifiers, and support vector machines (SVMs). Hyperparameters for each architecture are first manually tuned. Next, the best-performing model of each architecture is retrained on an impedance bandwidth subset in attempt to reduce dimensionality required for classification. Finally, Bayesian optimization is performed on each architecture to determine the best-performing hyperparameters. Results demonstrate that lower-frequency bandwidths are most sensitive to damage. The best-performing optimized model is a KNN model that attains an accuracy of 76.67%, an average AUC score of 81.85%, an overall performance score of 80.03%, as well as macro-recall and macro-F1 scores of 75.33% and 75%, respectively.

For the past three decades, total knee replacement has become the main solution for progressed knee injuries and diseases. Due to a lack of postoperative in vivo data, a universal correlation between intra- and postoperative soft tissue balance in the knee joint has not been established. In this work, an instrumented knee implant design with six piezoelectric transducers embedded in the tibial bearing is proposed. The aim of the presented device is to measure the total and compartmental forces as well as to track the location of contact points on the medial and lateral compartments of the bearing. A numerical analysis using finite element software is first performed to obtain the best sensory system arrangement inside the bearing. The chosen design is then used to fabricate a prototype of the device. Several experiments are designed and performed using the prototype, and the ability of the proposed system to track the location and magnitude of applied compartmental forces on the bearing is evaluated. The experimental results show that the instrumented knee bearing is able to accurately measure the compartmental force quantities with a maximum error of 2.6% of the peak axial load, and the CP locations with a maximum error of less than 1 mm.