Clinical Scorecard: Comparative Diagnostic Performance of Deep Learning Algorithms and Human Evaluators in Identifying Osteoporotic Vertebral Compression Fractures on Routine CT Imaging

At a Glance

Key Highlights

• Fragility fractures significantly increase morbidity, mortality, and healthcare costs, with high risk of subsequent fractures.
• CT imaging is valuable for detecting osteoporotic fractures but diagnosis is challenging, especially for mild fractures and in non-expert readers.
• Deep learning algorithms improve diagnostic accuracy and efficiency by detecting subtle fracture patterns beyond human capability.

Guideline-Based Recommendations

Diagnosis

• Use CT imaging with thin-slice reconstructions and bone kernel for detailed assessment of vertebral fractures.
• Apply semiquantitative Genant scale (grades 1–3) for fracture classification.
• Include fragility fractures in radiologic reports to prompt early treatment initiation.

Management

• Early identification of vertebral fractures to initiate appropriate osteoporosis treatment and fracture prevention strategies.

Monitoring & Follow-up

• Regular imaging follow-up may be indicated to monitor fracture progression or new fractures, especially in high-risk patients.

Risks

• Misclassification of mild fractures due to degenerative or developmental anomalies can lead to underdiagnosis.
• Variability in human assessment highlights need for standardized evaluation or AI assistance.

Patient & Prescribing Data

Patients aged >45 years undergoing routine CT scans with thoracic and lumbar spine coverage

Early and accurate fracture detection via CT and AI-assisted diagnosis can facilitate timely osteoporosis management and reduce subsequent fracture risk.

Clinical Best Practices

• Utilize deep learning algorithms alongside human evaluation to improve diagnostic accuracy of vertebral fractures on CT.
• Ensure CT protocols include thin-slice, high-resolution bone kernel reconstructions for optimal fracture visualization.
• Incorporate fracture assessment using the Genant scale at patient and vertebral levels, considering spinal region variability.
• Maintain awareness of differential diagnoses such as degenerative changes and developmental anomalies to avoid misinterpretation.
• Use multi-scanner, heterogeneous datasets for training AI models to enhance generalizability across clinical settings.

References

• Fragility fractures impact and risk
• Genant semiquantitative scale for vertebral fracture assessment
• VerSe datasets for vertebral fracture evaluation