Machine Learning and Artificial Intelligence Models for Fracture Detection in the Lower Limb on Plain Radiographs - An Umbrella Review of Systematic Reviews and Meta-Analyses

Artificial intelligence (AI) and machine learning (ML) systems -predominantly deep convolutional neural networks (CNN), and more recently vision transformers (ViT) and vision–language models have been applied extensively to fracture detection on plain radiographs. As primary studies have proliferated, numerous systematic reviews and meta-analyses have appeared, each pooling overlapping sets of primary studies and reaching conclusions that are not always concordant. The lower limb comprising the hip and proximal femur, femoral shaft and distal femur, knee and patella, tibia and fibula, ankle, and foot is clinically important and heterogeneous: some regions such as the hip and ankle are comparatively well studied, whereas others such as the foot, patella, and isolated fibula fractures are under-represented in the review-level literature. An umbrella review - a systematic synthesis of existing systematic reviews rather than of primary studies is the appropriate design to map this fragmented evidence base, characterise its consistency, quantify overlap between reviews, and appraise the methodological quality of the reviews themselves. This protocol specifies, in advance, the objectives, eligibility criteria, search strategy, appraisal instruments, data-extraction framework, overlap assessment, and synthesis plan, in order to minimise post-hoc decision-making and to support transparent, reproducible conduct. Status at registration: searches are complete (August to October 2026); title and abstract screening is in progress, having commenced on 16 October 2026; full-text screening and data extraction have not begun. This registration is therefore submitted for research that is underway, with the current state disclosed.