The Potential Clinical Utility of an Artificial Intelligence Model for Identification of Vertebral Compression...

Area of Science:

• Radiology
• Artificial Intelligence
• Osteoporosis Research

Background:

• Vertebral compression fractures (VCFs) are common, often linked to osteoporosis, and frequently underdiagnosed.
• Chest radiographs are routinely performed, offering a potential screening opportunity for VCFs.

Purpose of the Study:

• To evaluate the Annalise Enterprise CXR Triage Trauma AI model's accuracy in detecting VCFs on chest radiographs.
• To assess the AI model's utility in identifying patients with potential undiagnosed osteoporosis.

Main Methods:

• Retrospective analysis of 596 chest radiographs from four US hospitals (2015-2021).
• Consensus review by thoracic radiologists to identify VCFs.
• AI model inference and subsequent chart review for osteoporosis diagnoses and medication use.

Main Results:

• The AI model achieved high performance with an area under the receiver operating characteristic curve of 0.955.
• Sensitivity was 89.3% and specificity was 89.2% for VCF detection.
• A significant proportion of patients with AI-identified VCFs lacked a VCF diagnosis or osteoporosis treatment.

Conclusions:

• The AI model accurately identifies VCFs on chest radiographs.
• Automated VCF detection by AI can aid in identifying patients with undiagnosed osteoporosis.
• This AI application may facilitate timely initiation of disease-modifying osteoporosis medications.