Objectives <p>To explore the value of habitat radiomics analysis using multi-b-value diffusion weighted imaging in differentiating sinonasal small round cell malignant tumors (SRCMTs) from non-small round cell malignant tumors (non-SRCMTs).</p> Methods <p>This study enrolled 110 patients of sinonasal SRCMTs and non-SRCMTs, who were randomly divided into a training dataset (<i>n</i> = 88) and a test dataset (<i>n</i> = 22). Based on the true diffusion coefficient (Dt), perfusion fraction (f), and mean kurtosis coefficient (MK), which respectively characterize cellular density, perfusion, and heterogeneity, the sinonasal tumors were classified into four distinct habitats through k-means clustering. After extracting radiomics features and reducing dimensions, the radiomics, habitat, and fusion models were developed using Logistic regression (LR), NaiveBayes (NB), Support vector machine (SVM), RandomForest (RF), XGBoost (XGB), and MultiLayer perceptron (MLP) classifiers. Model performance was evaluated with the receiver operator characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA).</p> Results <p>The fusion model, specifically MK+Habitat_3_RF, which incorporated radiomics and habitat features, achieved the best performance with AUCs of 0.936 and 0.940 in the training and test datasets, respectively. The calibration curve and DCA results indicated improved fit and net benefit of the MK+Habitat_3_RF. Moreover, the AUC of the MK+Habitat_3_RF was significantly higher than the radiomics model of conventional DWI, the first-order feature “original_firstorder_Mean” of ADC (ADCmean), and visual assessment of two radiologists in the test dataset (<i>Ps</i> &lt; 0.05).</p> Conclusions <p>The MK+Habitat_3_RF can serve as an accurate and non-invasive tool for distinguishing between sinonasal SRCMTs and non-SRCMTs.</p>