Back to all papers

Machine learning and deep learning for ankle fracture detection in radiographs: A systematic review with narrative synthesis on diagnostic performance and future clinical integration.

May 25, 2026pubmed logopapers

Authors

Mangwani J,Jithin SP,Akram N,Ravi L,Vaishya R

Affiliations (5)

  • University Hospitals of Leicester NHS Trust, Leicester, United Kingdom; Centre for Digital Health and Precision Medicine (CDHPM), The Apollo University, Chittoor - 517127, Andhra Pradesh, India; Centre for Bioengineering and Biomedical Technologies (CBio), University of Bath, Bath, UK. Electronic address: [email protected].
  • Centre for Digital Health and Precision Medicine (CDHPM), The Apollo University, Chittoor - 517127, Andhra Pradesh, India. Electronic address: [email protected].
  • Royal Berkshire Hospital, Reading, United Kingdom. Electronic address: [email protected].
  • Centre for Digital Health and Precision Medicine (CDHPM), The Apollo University, Chittoor - 517127, Andhra Pradesh, India. Electronic address: [email protected].
  • Centre for Digital Health and Precision Medicine (CDHPM), The Apollo University, Chittoor - 517127, Andhra Pradesh, India; Indraprastha Apollo Hospital, New Delhi, India. Electronic address: [email protected].

Abstract

Ankle fractures are a prevalent orthopedic injury with X-rays being the primary diagnostic tool. Errors in radiological interpretation leads to missed fractures and delayed diagnosis. This systematic review with narrative synthesis aims to map the current landscape on the usage of Machine Learning (ML) and Deep Learning (DL) models for detecting ankle fractures from radiographs, with critical evaluation on the key trends and highlighting gaps in research. Following PRISMA guidelines, a systematic search was conducted on PubMed, Google Scholar, MEDLINE and Embase. Studies published between 2019 and June 2025 were screened, and relevant data, including model architectures, imaging modalities, dataset characteristics and performance metrics, were extracted and analyzed. A total of 27 studies met the inclusion criteria. The review reveals a clear trend of increasing research volume, with a focus on X-ray imaging and a preference for ResNet and InceptionV3 models. A significant evolution was observed towards more complex, clinically relevant tasks such as fracture localization. Some of the recent works introduced external validation of ankle-specific models, 3D-CT CNNs, and real-world reader-aid/workflow studies that demonstrate improved clinician accuracy and efficiency. However, a significant limitation across studies was a lack of heterogeneity in datasets and lack of external validation, which hindered generalizability. Machine learning holds significant promise as a diagnostic aid for ankle fracture detection, offering the potential to improve diagnostic accuracy. The addition of external validation, CT-based pipelines and multi-reader studies suggests a shift toward clinical integration. However, clinical application remains in the early stages. To bridge the gap between development and clinical application, future research must prioritize rigorous, multi-site validation studies and address practical considerations such as integration and regulatory approval.

Topics

Journal ArticleReview

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.