Morphological Sinus Features and Chronic Rhinosinusitis Risk: A Radiomics Study

ABSTRACT Objectives Anatomical factors of the sinuses significantly impact the progression of chronic rhinosinusitis (CRS). This study aims to investigate the correlation between morphological sinus features and the risk of CRS onset to provide evidence‐based support for clinical management. Methods This multicenter study enrolled 716 CRS patients and 115 healthy individuals from January 2019 to December 2023. CRS patients were categorized into eosinophilic CRS (eCRS) and non‐eosinophilic CRS (NeCRS) based on pathological results. The deep learning model nnU‐Net v2 was trained for automated segmentation of sinuses. Morphological features were extracted and then screened using PyRadiomics. Subsequently, logistic regression, random forest, and multi‐layer perceptron (MLP) were used to build CRS risk and subgroup evaluation models. Results CRS patients exhibited larger sinus volumes, surface areas, and multiple 2D/3D diameters compared to healthy individuals, while sphericity of maxillary and sphenoid sinuses was lower. A CRS risk prediction model was built using 20 morphological indicators. In the test cohort, the AUC values for the logistic regression, random forest, and MLP models were 0.884 (95% CI: 0.818–0.938), 0.890 (95% CI: 0.819–0.963), and 0.888 (95% CI: 0.822–0.952), respectively. Subgroup analysis showed that, although 18 indicators were selected for predicting the risk of CRS subtypes, the performance of all three models was relatively poor. Conclusions Multiple morphological sinus features are closely associated with the risk of CRS and could serve as significant indicators for assessing CRS risk. However, there are no significant differences in sinus morphology between eCRS and NeCRS patients.