# Computer Vision in Upper Limb Orthopaedics: A Scoping Review of Imaging-Based Algorithms for Fracture Detection and Radiographic Measurement

## TL;DR

This review explores how computer vision is being used in upper limb orthopaedics for tasks like fracture detection and radiographic measurement, highlighting its potential and current limitations.

## Contribution

The paper provides a scoping review mapping the current state of computer vision applications in upper limb orthopaedics, identifying key trends and gaps.

## Key findings

- Most studies focused on wrist and shoulder imaging using radiographs for fracture detection and postoperative assessment.
- Commonly used methods include convolutional neural networks, YOLO, and U-Net, showing high performance in classification and measurement tasks.
- Clinical adoption is limited due to small datasets, lack of external validation, and inconsistent reporting.

## Abstract

Computer vision techniques are increasingly applied to medical imaging and may provide valuable assistance in upper limb orthopaedics, a field where radiographs, CT, MRI, and ultrasound are central to diagnosis and treatment planning. The development of deep learning has made automated interpretation of orthopaedic imaging both feasible and increasingly accurate. This review maps and characterises the current use of computer vision in upper limb orthopaedics, describing the clinical problems addressed, the imaging modalities used, and the methodological approaches reported in the literature.

Ovid MEDLINE and Embase were searched from January 1995 to October 2025. Studies were eligible if they applied an automated or semi-automated computer vision technique to upper limb imaging for diagnostic or planning purposes. Conference abstracts and non-orthopaedic applications were excluded. Two reviewers independently screened titles, abstracts, and full texts. Data extraction followed the PRISMA extension for scoping reviews (PRISMA-ScR) framework.

Sixteen studies met the inclusion criteria. Most focused on wrist and shoulder imaging, particularly radiograph-based fracture detection and postoperative morphometric assessment. Convolutional neural networks, detection networks (e.g., YOLO), and segmentation architectures (e.g., U-Net) were most commonly used. Performance was generally high across fracture classification and measurement tasks, although most studies were retrospective, single-centre-based, and lacked external validation.

Computer vision in upper limb orthopaedics remains at an early but promising stage. Automated systems for fracture detection and radiographic measurement demonstrate encouraging accuracy, yet widespread clinical use is limited by small datasets, lack of prospective validation, and inconsistent reporting. Future research should prioritise reproducible multicentre studies, inclusion of soft-tissue modalities, and exploration of intra-operative and real-time clinical applications.