OsteoAI
Incidental findings of VCFs in CTs using AI.
Modality:
CT
Pathology:
Vertebral compression fractures
Status:
Current
CSC Lead: Elise
Osteoporosis is easily treatable if identified but can lead to complex fractures at other sites if undiagnosed, especially in the hip, where mortality is over 50% at 12 months for individuals over the age of 80. Currently, 70% of vertebral compression fractures (VCFs) are undiagnosed, even though they are the most common osteoporotic fracture. This project works in conjunction with the current Fracture Liaison Service quality improvement project.
Clinical Lead: Davina Mak
Rationale
Among patients with vertebral compression fractures, 77% of scans are accurately identified, and 43% recommend referrals. Both of these metrics could be improved with the use of AI tools. Timely identification of vertebral compression fractures, which indicate osteoporosis, allows clinicians to refer patients for osteoporosis treatment, preventing complex fractures in vulnerable patients.
Training Data
The training data for this project consists of a few thousand of CTs where VCFs have been identified. The inclusion’ and exclusion criteria have been strictly followed and relate to patient’s age and clinical pathway.
Testing Data
Testing data consists of a few hundred CT scans where spine is visible in the CT. The dataset is separate from the training dataset.
Risks
False positives leading to an increase in workload for FLS clinic beyond its capacity. Patients on oncological pathways being referred where the VCF are the result of cancer treatment and not osteoporosis.
Goals
Increase the rates of identification and treatment referrals after detection to reduce the occurrence of complex fractures at other sites.
Success Criteria
Accurate referral of all incidentally found VCFs to FLS clinic.
Alternatives
Commercial products for this clinical problem exist. One was explored for approximately a year but was eventually withdrawn due to a number of contracting complications.