Validation of a Deep Learning Tool for Detection of Incidental Vertebral Compression Fractures
Michelle Dai1,2, Bryan-Clement Tiu1, Jacob Schlossman1
1Irvine School of Medicine, University of California, Irvine, CA.
Journal of Computer Assisted Tomography
|January 29, 2025
Summary
A deep learning tool accurately detected incidental vertebral compression fractures (VCF) in CT scans, outperforming clinical reports. This multicenter study validates the tool's generalizability and clinical applicability for VCF detection.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Vertebral compression fractures (VCFs) are common, particularly in older adults.
- Incidental VCFs detected on CT scans can be missed or misdiagnosed.
- Automated detection tools can improve diagnostic accuracy and efficiency.
Purpose of the Study:
- To evaluate the performance of a deep learning algorithm (CINA-VCF) for detecting incidental VCFs.
- To validate the CINA-VCF tool across multiple institutions and imaging vendors.
Main Methods:
- Retrospective, multicenter, multinational blinded study of anonymized CT scans (n=474) from patients ≥50 years old.
- Images processed by CINA-VCF v1.0; ground truth established by consensus of 3 radiologists.
- Performance analyzed overall, by VCF severity, and compared to clinical radiology reports.
Main Results:
- CINA-VCF achieved high performance with an AUC of 0.97, accuracy 93.7%, sensitivity 95.2%, and specificity 92.9%.
- Excellent sensitivity for severe VCFs (grades 2-3: 89.7%-99.0%); specificity varied for less severe fractures.
- The tool outperformed existing clinical radiology reports in VCF detection.
Conclusions:
- The CINA-VCF deep learning tool demonstrates high accuracy and generalizability for detecting incidental VCFs.
- The tool's performance remained consistent across various subgroups, supporting clinical utility.
- Despite limitations with confounding pathologies, the study validates CINA-VCF for real-world clinical settings.
