Related Experiment Video
Updated: Jun 30, 2026
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Radiomics to discriminate between axial spondyloarthritis and axial psoriatic arthritis and to predict TNFi therapy
Vincenzo Venerito1, Sergio Del Vescovo1, Crescenzio Scioscia1
1Rheumatology Unit-Department of Precision and Regenerative Medicine and Ionian Area, University of Bari, Bari, Italy.
Objectives:
Axial spondyloarthritis (axSpA) and axial psoriatic arthritis (axPsA) represent entities with ongoing debate regarding their classification as distinct diseases or variants of the same condition. Although magnetic resonance imaging (MRI) is instrumental in diagnosing, traditional assessment may not fully capture subtle differences between them. This study aims to investigate whether radiomic features extracted from sacroiliac joint (SIJ) bone marrow oedema (BME) on MRI can discriminate between axSpA and axPsA, and predict tumour necrosis factor inhibitor (TNFi) therapy persistence in patients with axSpA.
Methods:
We included patients who underwent an SIJ MRI at our hospital. BME regions were segmented on short-tau inversion-recovery sequences, and 120 standardised radiomic features were extracted using PyRadiomics. For diagnostic discrimination, an XGBoost algorithm was employed and evaluated through 5-fold cross-validation. For patients with axSpA who initiated TNFi therapy, an exploratory Cox regression analysis was performed to identify radiomic predictors of treatment persistence.
Results:
We analysed MRI scans from 66 patients (41 axSpA, 25 axPsA). The XGBoost classifier achieved an accuracy of 0.80 ± 0.07 and area under the receiver operating characteristic curve of 0.77 ± 0.09 in differentiating axSpA from axPsA. The most discriminative features included texture parameters, shape characteristics, and grey-level intensity distributions. For TNFi persistence, multivariate Cox regression identified 2 shape-based features as independent predictors: increased sphericity, associated with discontinuation risk (hazard ratio [HR] 2.10, 95% CI: 1.09-4.05), while increased elongation showed a protective effect (HR 0.50, 95% CI: 0.25-0.99).
Conclusions:
This proof-of-concept study suggests radiomic SIJ BME features may help differentiate axSpA and axPsA and could be associated with TNFi persistence in patients with axSpA. These preliminary findings suggest that these conditions may exhibit distinct radiomic phenotypes.