Spondyloarthritis & psoriatic arthritis
Machine Learning Pipeline for Predicting Bone Marrow Edema Along the Sacroiliac Joints on Magnetic Resonance Imaging
Authors' conclusion
We propose a fully automated ML pipeline that enables objective and standardized evaluation of BME along the SI joints on MRI.
Study context
- Population
- Patients with spondyloarthritis (SpA), postpartum individuals and healthy subjects with sacroiliac (SI) joint magnetic resonance imaging (MRI) (semicoronal T1/T2-weighted)
- Comparison
- Cross-validation vs Test data set
- Test data setn = 243
- Outcome
- Prediction of bone marrow edema (BME) on a quadrant level in SI joint MRI, vs ground truth by consensus of human readers
Methodological detailsfully automated machine learning pipeline, ResNet18 backbone, five-fold cross-validation, independent test data set
- Fully automated machine learning pipeline
- ResNet18 backbone
- Five-fold cross-validation
- Independent test data set
Performance
As reported in the abstract
Quadrant-level BME prediction
| Test or model | AUC | Balanced accuracy | F1 score |
|---|---|---|---|
| Cross-validation | 94.5% | 80.5% | 64.1% |
| Test data set | 88.2% | 72.1% | 50.8% |
AUC
Cross-validation94.5%
Test data set88.2%
Balanced accuracy
Cross-validation80.5%
Test data set72.1%
F1 score
Cross-validation64.1%
Test data set50.8%
Further findings
Patient-level BME prediction (≥1 positive quadrant)
Balanced accuracy
SI joint detection and ilium/sacrum segmentation
- SI joint detection precision
- 98.4%
- Ilium segmentation intersection over union
- 85.6%
- Sacrum segmentation intersection over union
- 67.9%
Original publication
Machine Learning Pipeline for Predicting Bone Marrow Edema Along the Sacroiliac Joints on Magnetic Resonance Imaging. Arthritis Rheumatol. Published 10 November 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →