External validation of a convolutional neural network algorithm for opportunistically detecting vertebral fractures in routine CT scans
Authors' conclusion
The algorithm demonstrated excellent performance in the identification of vertebral fractures in a cohort of chest and abdominal CT scans of Chinese patients ≥50 years.
Study context
- Population
- Men and women aged ≥50 years with CT scans of the chest, abdomen and thoracolumbar spine from two Chinese studies and clinical data; median age 62 years (IQR 56–67), 2,654 (55.2%) females628 scans (13.1%) with ≥1 VF
- Participants
- 4,810 men
- Comparison
- Convolutional Neural Network (CNN) model vs reference standard readings
- Outcome
- CNN performance in identifying scans with ≥1 moderate or severe VF (grade 2-3)
Methodological detailsexternal validation, retrospective, blinded to clinical information, Scans acquired January 2013 to January 2019 on 16 different CT scanners from three manufacturers
- External validation
- Retrospective
- Blinded to clinical information
- Scans acquired January 2013 to January 2019 on 16 different CT scanners from three manufacturers
Performance
As reported in the abstract
Identification of scans with ≥1 moderate or severe VF
| Test or model | AUROC | Accuracy | Kappa | Sensitivity | Specificity |
|---|---|---|---|---|---|
| CNN | 0.94 (95% CI: 0.93-0.95) | 93% (95% CI: 93%-94%) | 0.75 (95% CI: 0.72-0.77) | 94% (95% CI: 92-96%) | 93% (95% CI: 93-94%) |
CNN
AUROC0.94 (95% CI: 0.93-0.95)
Accuracy93% (95% CI: 93%-94%)
Kappa0.75 (95% CI: 0.72-0.77)
Sensitivity94% (95% CI: 92-96%)
Specificity93% (95% CI: 93-94%)
Original publication
External validation of a convolutional neural network algorithm for opportunistically detecting vertebral fractures in routine CT scans. Osteoporos Int. Published 7 September 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →