Project Overview

Osteoporotic vertebral fractures are a leading cause of disability and spinal deformity in ageing populations. While some fractures heal without progression, others lead to severe kyphosis and chronic pain. Current tools like FRAX do not account for vertebra-specific features or deformity progression. This project addresses the clinical gap by developing AI models that predict progression based on radiographic and demographic information.

Objectives

  • Build deep learning models that analyse spine radiographs to assess vertebral morphology and deformation risk.
  • Incorporate clinical data such as age, prior fractures, and bone health indicators to enhance predictive power.
  • Provide interpretable outputs to assist clinicians in making proactive decisions.

Methodology

  • Data Source: Spine radiographs and clinical records from Bedfordshire NHS Foundation Trust and public imaging datasets.
  • Annotation: Vertebral height loss, shape metrics, and alignment angles will be labelled for model training.
  • Model Development: CNN-based architectures with temporal components to evaluate deformity progression over time.
  • Validation: Compare AI predictions against outcomes and traditional fracture risk calculators.

Clinical Impact

The tool is expected to:

  • Improve early identification of high-risk patients
  • Inform decisions on bracing, monitoring, or early intervention
  • Reduce long-term disability and healthcare burden from progressive deformities

Collaborators