Opportunistic Identification of Vertebral Compression Fractures on CT Scans of the Chest and Abdomen, Using an AI...

Area of Science:

• Radiology
• Medical Imaging
• Osteoporosis Research

Background:

• Vertebral compression fractures (VCFs) are a significant health concern.
• Automated detection algorithms aim to improve diagnostic efficiency.
• Real-world performance evaluation is crucial for clinical implementation.

Purpose of the Study:

• To evaluate the real-life performance of the HealthVCF algorithm for detecting moderate and severe VCFs.
• To assess the impact of the algorithm on patient treatment and diagnostic workflow.
• To determine the generalizability of the HealthVCF algorithm to a Danish population.

Main Methods:

• Approximately 10,000 CT scans were processed by the HealthVCF algorithm.
• A subset of 1000 CT scans was manually reviewed by specialized radiographers for performance assessment.
• A retrospective cohort of 538 patients with VCFs was analyzed at baseline and 6-month follow-up.

Main Results:

• The HealthVCF algorithm demonstrated a sensitivity of 0.68 and specificity of 0.91.
• A higher mortality rate was observed in patients not previously diagnosed with osteoporosis.
• Patients on bisphosphonates showed a lower mortality rate.
• The algorithm's performance was poorer than anticipated and not generalizable to the Danish population.

Conclusions:

• The HealthVCF algorithm, despite limitations, can aid in prioritizing VCF identification due to its specificity.
• Integration with Fracture Liaison Services (FLS) is recommended to enhance efficiency and increase osteoporosis diagnoses.
• The current implementation showed limited impact on new osteoporosis diagnoses and DXA referrals within 6 months.