Fat infiltration in the sacroiliac joint is an important feature in the diagnosis and progression of axial spondyloarthritis (axSpA)，yet its quantitative assessment and metabolic underpinnings remain poorly characterized. This study aimed to evaluate the performance of MRI-based radiomic quantification of fat infiltration using Dixon sequences and to characterize the serum lipidomic profile in patients with axSpA.

In this prospective cross-sectional study, Dixon sequence MRI (a fat suppression technique based on water-fat separation) of the sacroiliac joints was acquired from consecutive patients presenting with back pain between January 2023 and December 2024. We extracted quantitative radiomics features from the fat-only images. Unsupervised clustering was then used to identify and characterize imaging-based patient clusters. Additionally, fasting plasma samples were collected for lipidomic profiling using ultra-high-performance liquid chromatography coupled with mass spectrometry (HPLC-MS).

A total of 95 patients (55 men and 40 women) were enrolled in the study. Two clinically distinct patient clusters were identified based on quantitative fat imaging data. Regarding disease classification, Cluster 1 contained significantly more patients with radiographic axSpA (r-axSpA) (80% vs.20%, P < 0.001), whereas Cluster 2 contained more patients with non-radiographic axSpA (nr-axSpA) and healthy controls (29% vs. 71%, P < 0.001). Patients in Cluster 1 had more advanced sacroiliac joint damage grading, with a higher proportion of grades II, III, and IV and a lower proportion of grades 0 and I compared with Cluster 2 (P < 0.001). Cluster 1 also had a higher proportion of male patients (78% vs. 22%, P < 0.001). Patients in Cluster 1 were older (39.83 ± 13.99 vs. 32.34 ± 8.33 years, P=0.002) and had a longer disease duration (79.28 ±78.87 vs. 48.99±53.65 months, P=0.03) than those in Cluster 2. In terms of bone turnover markers (BTMs), Cluster 1 exhibited significantly higher levels of N-terminal mid-fragment osteocalcin (N-MID) (P＜0.001), total procollagen type 1 N-terminal propeptide (t-P1NP) (P=0.007), β-isomerized C-terminal telopeptide of type I collagen (β-CTX) (P=0.004), calcitonin (CT) (P=0.002), and alkaline phosphatase (AKP) (P=0.002), indicating increased bone remodeling activity in this cluster. Between the two clusters, 45 lipids (positive ion mode) and 22 lipids (negative ion mode) showed significant changes. Lipidomic analysis revealed distinct metabolic signatures between the two clusters. In positive ion mode, 45 significantly altered lipids were enriched in glycerophospholipid and sphingolipid metabolism, necroptosis, and cancer-related choline metabolism. In negative ion mode, 22 significantly altered lipids were enriched in insulin resistance, thermogenesis, and diabetic complications-related pathways.

MRI-based radiomic quantification of sacroiliac joint fat infiltration on Dixon images enables objective phenotyping of axial spondyloarthritis and identifies a subgroup with more advanced structural damage, increased bone remodeling activity, and a distinct circulating lipidomic profile. These findings support quantitative fat imaging as a noninvasive biomarker of disease heterogeneity and suggest that dysregulated lipid metabolism may be linked to fat metaplasia–related structural remodeling in axSpA. Integrating Dixon-derived radiomics with lipidomics may improve patient stratification and provide further insight into the metabolic basis of disease progression, although longitudinal validation in larger cohorts is still warranted.