Joint Track Machine Learning: An autonomous method for measuring 6DOF TKA kinematics from single-plane x-ray images
Dynamic radiographic measurement of 3D TKA kinematics has provided important information for implant design and surgical technique for over 30 years. However, current methods of measuring TKA kinematics are too cumbersome or time-consuming for practical clinical application. Even state-of-the-art techniques require human-supervised initialization or human supervision throughout the entire optimization process. Elimination of human supervision could potentially bring this technology into clinical practicality. Therefore, we propose a fully autonomous pipeline for quantifying TKA kinematics from single-plane imaging. First, a convolutional neural network segments the femoral and tibial implants from the image. Second, segmented images are compared to Normalized Fourier Descriptor shape libraries for initial pose estimates. Lastly, a Lipschitzian optimization routine minimizes the difference between the segmented image and the projected implant. This technique reliably reproduces human-supervised kinematics measurements from internal datasets and external validation studies, with RMS differences of less than 0.7mm and 4{\deg} for internal studies and 0.8mm and 1.7{\deg} for external validation studies. This performance indicates that it will soon be practical to perform these measurements in a clinical setting.
PDF Abstract