A study on whether deep learning models based on CT images for bone density classification and prediction can be used for opportunistic osteoporosis screening
Authors' conclusion
Using deep learning neural networks to process opportunistic CT scan images of the body can accurately predict bone density values and perform bone density three-classification diagnosis
Study context
- Population
- Cases of opportunistic CT scans with QCT results as reference standard
- Participants
- 1,219 opportunistic CT scans
- Comparison
- Deep learning models on opportunistic CT vs QCT reference standard
- Training setn = 703
- Test setn = 176
- Independent test setn = 340
- Outcome
- Three-category classification (normal, osteopenia, osteoporosis), mean AUC
Methodological detailsdeep learning (VB-Net segmentation, DenseNet classification and regression), QCT reference standard, independent test set from 3 hospitals and 4 CT scanners
- Deep learning (VB-Net segmentation, DenseNet classification and regression)
- QCT reference standard
- Independent test set from 3 hospitals and 4 CT scanners
Performance
As reported in the abstract
| Test or model | Mean AUC | Specificity | F1 score | Accuracy | Precision | Recall |
|---|---|---|---|---|---|---|
| Training set | 0.999 | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract | — not stated in the abstract |
| Test set | 0.970 | 0.956 | 0.903 | 0.909 | 0.899 | 0.908 |
| Independent test set | 0.933 | 0.899 | 0.798 | 0.815 | 0.792 | 0.81 |
Mean AUC
Training set0.999
Test set0.970
Independent test set0.933
Specificity
Training set— not stated in the abstract
Test set0.956
Independent test set0.899
F1 score
Training set— not stated in the abstract
Test set0.903
Independent test set0.798
Accuracy
Training set— not stated in the abstract
Test set0.909
Independent test set0.815
Precision
Training set— not stated in the abstract
Test set0.899
Independent test set0.792
Recall
Training set— not stated in the abstract
Test set0.908
Independent test set0.81
— not stated in the abstract
Further findings
| Outcome | Training set | Test set | Independent test set |
|---|---|---|---|
| Bone density prediction regression model | |||
| MAE | 3.15 | 6.303 | 10.257 |
| RMSE | 4.127 | 8.561 | 13.507 |
| R-squared | 0.991 | 0.962 | 0.878 |
| Pearson correlation coefficient | 0.996 | 0.981 | 0.94 |
- Bone density prediction regression model: p values all < 0.001
Bone density prediction regression model
- Predicted values highly positively correlated with bone density values; significant linear relationship
Original publication
A study on whether deep learning models based on CT images for bone density classification and prediction can be used for opportunistic osteoporosis screening. Osteoporos Int. Published 5 September 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →