An Ultrasound-Based Method to Measure Knee Kinematics Enabled by Deep Learning

Description

This presentation from the ISTA 2024 conference held in Nashville, led by PhD student Matthew Blomquist, discusses a study aimed at improving the assessment of knee arthroplasty kinematics in clinical settings. Tracking secondary kinematics, such as anterior, posterior, and varus/valgus movements, is traditionally done through subjective manual assessments, presenting challenges in accuracy. The presentation highlights the limitations of current methods, including thermometers and stress radiographs, which either lack accuracy due to soft tissue compliance or are not easily used in clinical environments.



The study proposes an innovative use of ultrasound technology combined with deep learning techniques to enhance measurement accuracy and accessibility. The methodology involves training a convolutional neural network with kinematic data from robotic laxity tests across multiple specimens and knee flexion angles. Results indicate that this deep learning approach yields root mean square errors of less than 1.9 degrees for anterior-posterior laxity and less than 1.2 degrees for varus/valgus movements, demonstrating promise for real-time, automated knee kinematics measurements in clinical practice.



Blomquist concludes that the integration of ultrasound with deep learning could significantly facilitate clinical decision-making by providing faster and more accurate evaluations of knee joint function.

Specialties