Zhihao Zhang, Zongliang Huang, Yandan Shi, Jinhua Hu, Peng Lei, Jiayu Wu, Ziyuan Yang, Jinqing Su, Qi Wu, Haizhi Ma, Changhui Huang, Xiqi Zhu
By quantifying BME spatial heterogeneity, the multi-habitat model provides an objective imaging framework to characterize active sacroiliitis. This objective framework offers high spatial interpretability, acting as an innovative tool for quantifying BME spatial heterogeneity, which may serve as a foundation for future studies exploring its role in assessing clinical disease activity. Key Points • Multi-habitat radiomics on T2-FS MRI objectively quantifies the spatial heterogeneity of sacroiliac bone marrow edema. • The habitat-based SVM model significantly outperforms traditional whole-lesion radiomics in identifying early active sacroiliitis. • The peripheral transition zone and hyperintense core act as critical imaging drivers for axSpA diagnosis.
INTRODUCTION: Conventional radiomics neglects the spatial heterogeneity of bone marrow edema (BME). We developed a multi-habitat radiomics model using fat-suppressed T2-weighted (T2-FS) MRI to quantify BME spatial patterns in axial spondyloarthritis (axSpA) to objectively assist in the identification of active sacroiliitis.
METHODS: This multicenter study included 439 patients with suspected axSpA undergoing sacroiliac joint MRI (divided into train/validation/external test cohorts: n = 192/84/163). Unsupervised K-means clustering of voxel-wise T2-FS features partitioned BME into four habitats. Radiomics features were extracted from habitats and whole-lesions to optimize support vector machine (SVM) classifiers. A "Leave-One-Habitat-Out" ablation study evaluated each subregion's diagnostic contribution.
RESULTS: The multi-habitat SVM model achieved an area under the receiver operating characteristic curve (AUC) of 0.883 in the independent test set, significantly outperforming the whole-lesion model (AUC 0.714; P < 0.001), with a net reclassification improvement of 0.927. The ablation study identified the putative peripheral transition zone (H1) and hyperintense core (H2) as primary radiological diagnostic drivers; excluding H1 caused the most severe performance degradation (AUC decreased to 0.644, P < 0.001).
CONCLUSION: By quantifying BME spatial heterogeneity, the multi-habitat model provides an objective imaging framework to characterize active sacroiliitis. This objective framework offers high spatial interpretability, acting as an innovative tool for quantifying BME spatial heterogeneity, which may serve as a foundation for future studies exploring its role in assessing clinical disease activity. Key Points • Multi-habitat radiomics on T2-FS MRI objectively quantifies the spatial heterogeneity of sacroiliac bone marrow edema. • The habitat-based SVM model significantly outperforms traditional whole-lesion radiomics in identifying early active sacroiliitis. • The peripheral transition zone and hyperintense core act as critical imaging drivers for axSpA diagnosis.