Spondyloarthritis & psoriatic arthritis
High‐Resolution Deep Learning Dixon Magnetic Resonance Imaging of the Sacroiliac Joints Is Noninferior to Standard Magnetic Resonance Imaging in Patients With Suspected Axial Spondyloarthritis
Authors' conclusion
DL-Dixon provides clinically acceptable diagnostic performance while substantially reducing scan time.
Study context
- Population
- Patients with chronic low back pain and suspected axial spondyloarthritis (axSpA)19 diagnosed with axSpA
- Participants
- 76 patients
- Comparison
- High-resolution deep learning-reconstructed Dixon sequence vs multisequence standard MRI of the sacroiliac joints
- Outcome
- Diagnostic accuracy (AUC; balanced accuracy) vs clinical diagnosis; noninferiority margin 0.05
Methodological detailsnon-inferiority, three blinded readers, 3T MRI
- Non-inferiority
- Three blinded readers
- 3T MRI
Performance
As reported in the abstract
Diagnostic accuracy for axSpA (AUC)
| Test or model | AUC |
|---|---|
| DL-Dixon | 0.86 (95% CI 0.76-0.96) |
| Standard magnetic resonance imaging (sMRI) | 0.87 (95% CI 0.78-0.96) |
AUC
DL-Dixon0.86 (95% CI 0.76-0.96)
Standard magnetic resonance imaging (sMRI)0.87 (95% CI 0.78-0.96)
Further findings
| Outcome | DL-Dixon | sMRI |
|---|---|---|
| Interreader agreement (κ)Almost perfect for sMRI, substantial for DL-Dixon | 0.71 | 0.81 |
| Lesion-level performance of DL-Dixon vs sMRI reference (AUC) | ||
| AUC erosions | 0.83 | — not stated in the abstract |
| AUC fat metaplasia | 0.80 | — not stated in the abstract |
| AUC joint space changes | 0.80 | — not stated in the abstract |
| AUC osteitis | 0.73 | — not stated in the abstract |
| AUC sclerosis | 0.64 | — not stated in the abstract |
— not stated in the abstract
Scan time
- Scan time reduced with DL-Dixon
- 73%
Original publication
High‐Resolution Deep Learning Dixon Magnetic Resonance Imaging of the Sacroiliac Joints Is Noninferior to Standard Magnetic Resonance Imaging in Patients With Suspected Axial Spondyloarthritis. Arthritis Rheumatol. Published 21 April 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →