Today, total knee arthroplasty (TKA) is one of the most common surgeries in the United States and is expected to grow rapidly over the next 30 years. While many TKA procedures are successful, some implants fail after primary surgery, leading to severe knee pain and costly revision surgery. As the number of TKA surgeries grows, the number of implant failures also continues to rise. In order to reduce the percentage of patients who have to undergo revision surgery, there has been a search for early detection methods to identify and treat damaged implants. Previous work has demonstrated the use of structural health monitoring techniques for detecting mechanical failure in simulated TKA implants using electromechanical impedance techniques and machine learning algorithms. However, current methods lack the accuracy necessary for medical implementation and only consider classical machine learning methods. This work aims to expand on previous methods by implementing convolutional neural networks to detect aseptic loosening and aseptic debonding in simulated TKA implants using root-mean-square-difference impedance maps. The results of the experiments show that the algorithm is able to predict damage category, damage severity, and a combination of damage category and severity with an accuracy of 90.9%, 90.9%, and 83.3%, respectively, demonstrating improvements over previous methods.
|