Opportunistic CT Screening-Machine Learning Algorithm Identifies Majority of Vertebral Compression Fractures: A...

Area of Science:

• Radiology
• Artificial Intelligence
• Bone Health

Background:

• Vertebral compression fractures (VCF) are prevalent in individuals over 50 but frequently go undiagnosed.
• Early detection is crucial for managing osteoporosis and preventing further skeletal fragility.

Purpose of the Study:

• To evaluate the diagnostic performance of a machine learning-based algorithm for detecting VCFs in CT images.
• To assess the algorithm's sensitivity and specificity in identifying any VCF and moderate/severe VCF.

Main Methods:

• A blinded validation study was conducted using CT scans from 1087 participants (aged 50+) at a tertiary-care center.
• Two neuroradiologists independently evaluated scans for VCF, with disagreements resolved by a senior neuroradiologist.
• The VCF detection algorithm processed scans separately to estimate its diagnostic accuracy.

Main Results:

• The algorithm demonstrated a sensitivity of 0.66 and specificity of 0.90 for detecting any VCF.
• For moderate/severe VCFs, the algorithm achieved a sensitivity of 0.78 and specificity of 0.87.
• The algorithm was unable to evaluate scans for 113 out of 1200 participants.

Conclusions:

• The VCF detection algorithm shows potential for identifying patients at higher risk of fractures.
• Integration into radiology workflows could support osteoporosis diagnosis and timely therapeutic interventions.