Multicenter, Multinational, and Multivendor Validation of an Artificial Intelligence Application for Acute Cervical...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Previous AI studies for cervical spine fracture (CSFx) detection often lacked diverse data validation.
• Many studies focused solely on overall case-level classification, not detailed fracture analysis.

Purpose of the Study:

• To evaluate an AI application for acute CSFx detection.
• Assess AI performance in case-level classification, fracture localization, and spinal level labeling.
• Validate AI on multicenter, multinational, and multivendor CT data.

Main Methods:

• Retrospective collection of non-enhanced CT scans from US and French teleradiology companies and a US university hospital.
• Independent radiologist labeling of acute CSFx to establish a reference standard.
• Assessment of AI's per-case diagnostic performance, localization accuracy (bounding box PPV), and vertebral level labeling agreement.

Main Results:

• The AI achieved high diagnostic performance: 90.3% sensitivity, 91.9% specificity, 91.2% accuracy, and an AUC of 0.91.
• The AI demonstrated strong fracture localization with a per-bounding box PPV of 84.4%.
• Vertebral level labeling by the AI showed high agreement at 97.3%.

Conclusions:

• The AI application for acute CSFx detection exhibits high diagnostic accuracy.
• The AI performs well in fracture localization and precise spinal level identification.
• The AI's robust performance on diverse datasets supports its clinical utility.