3D Printed Ultrasound Cuff with Machine Learning Algorithms for the Detection of Knee Implant Loosening (tbd)

Abstract

Millions of patients worldwide undergo joint replacements each year, and in recent years, these procedures are being implemented in increasingly younger patients. However, based on the ages and general health of the patients, recipients of these replacements can outlive their original implant; thus, creating the need for revision surgeries. Revision surgery and recovery are painful and inconvenient for the patient. One of the main causes of revision surgery is implant loosening, which occurs when there is a space along the interface of the implant and the bone. Current methods to detect knee implant loosening include radiography, CT scans, and MRI imaging, which are costly and cannot detect early onset problems. Additionally, the diagnosis is solely based on the discretion of the clinician who is overlooking the imaging. We have proposed and built a prototype for a wearable cuff device that uses ultrasound signals to detect implant loosening of knee replacements. Our cuff can detect loosening at earlier stages at a lower cost than these current methods. Our ultrasound sensors, which are built into our 3D printed wearable, send and receive sinusoidal signals that can then be run through our predictive machine learning algorithms. The algorithm results help doctors identify different types of loosening of varying degrees of severity and locations around the implant. We successfully designed and 3D printed a flexible, wearable cuff that meets all of our engineering criteria. Additionally, we tested the feasibility of sending and receiving signals from the sensors within our cuff. We simulated the knee cross-section with and without knee implant defects with eight piezoelectric sensors applied symmetrically on the skin. The machine learning algorithms that we trained with our simulation data can sense the presence of a defect, and if there is a defect, identify details about location, size and shape of the loosening. In future experiments, our cuff will send and receive signals from knee implant-introduced cadaver knees with and without induced defects. We will run these data sets through our machine learning algorithms and test the entire product as a whole. Our cuff can be resized and our algorithms can be retrained on different joint interfaces in order to be applied to different joint replacement types (such as hip, spine, shoulder, etc).