Artificial intelligence-based detection of posterior tibial slope on X-ray images of unicompartmental knee arthroplasty patients

Abstract

ObjectiveThis paper aims to improve the assessment of the outcomes of unicompartmental knee arthroplasty (UKA) from the angle of posterior tibial slope (PTS) on X-Ray images, using the artificial intelligence (AI) tool of RetinaNet. MethodsFirstly, RetinaNet was trained and tested on patients who underwent unilateral knee replacement surgery in our hospital due to osteoarthritis in the medial compartment of either their left or right knee between July 2018 and July 2022. The trained network was applied to detect the region of interest (ROI) on the pre- and postoperative X-ray images of each subject. Next, the subjects were divided into three groups according to the PTS changes measured by the trained RetinaNet. Furthermore, the surgical effect of UKA was evaluated from multiple angles, including pre- and postoperative PTSs, knee joint mobility (KJM) values, the Hospital for Special Surgery (HSS) scores, as well as the Joint Replacement Forgetting Index (JRFI). ResultsAfter training, the RetinaNet achieved an astounding accuracy level, with the Cronbach's alpha value 0.864 (95%CI 0.762–0.915). There significant differences were found between Group II and Group I (P = 0.017) and Group III and Group I (P = 0.032); in terms of JRFI, Group II had a significantly better than Group I (P = 0.011) and Group III (P = 0.037). ConclusionsThe trained RetinaNet is suitable for assisting with PTS measurements in X-ray images; the PTS variation through UKA should be controlled within 2° to ensure the best possible surgical effect.