Abstract
Computer vision and image-based artificial intelligence (AI) are increasingly being used in orthopaedic imaging. Applications in lower limb orthopaedic surgery present unique diagnostic, biomechanical and surgical planning challenges. This review systematically maps how computer vision has been applied to lower limb orthopaedic imaging, identifying key tasks, modalities and algorithmic trends in published studies. A scoping review was conducted following Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines. Ovid MEDLINE and Embase were searched from January 1995 to October 2025. Studies were included if they applied an automated or semi-automated computer vision method to imaging of the hip, knee, ankle or foot. Exclusions included conference abstracts, ongoing clinical trials and studies without full text. Twenty studies met the inclusion criteria. The knee was the most frequently studied region (40%), followed by the hip (30%), ankle (15%) and foot (10%). Radiographs (40%) and CT (45%) were the dominant imaging modalities, while MRI (10%) and ultrasound (5%) were less common. Deep learning was employed in 85% of studies, primarily using convolutional architectures such as U-Net, YOLO, and ResNet. Across tasks, reported diagnostic accuracies were typically above 85%, and segmentation Dice coefficients frequently exceeded 0.85. Despite strong technical results, only 20% of studies included external validation, with no prospective clinical trials identified. Computer vision research in lower limb orthopaedics is diversifying and progressing from diagnostic classification toward quantitative measurement and intraoperative integration. Despite promising accuracy and automation potential, the evidence base is constrained by predominantly single-centre retrospective designs and scarce external validation. Future studies should prioritise reproducibility, large-scale validation and clinical practice deployment.