OBJECTIVE: Prediction of the therapeutic efficacy of uterine artery embolization (UAE) for adenomyosis (AM) using an MRI-based radiomics model combined with clinical characteristics. METHODS: A retrospective analysis was conducted on 126 patients with AM who underwent UAE at the Interventional Radiology Department of the Second Affiliated Hospital of Soochow University. Radiomics features were extracted from uterine lesions using axial T-weighted imaging with fat suppression (TWI-FS) sequences obtained prior to treatment. Following feature selection using the mRMR and LASSO algorithms, radiomics models were developed to predict the lesion necrosis rate in AM after UAE. These models employed the following classifiers: C-SVC, Nu-SVC, Logistic Regression (LR), Random Forest (RF), AdaBoost, and XGBoost. The optimal radiomics model was subsequently identified through receiver operating characteristic (ROC) curve analysis. Relevant clinical characteristics were screened using univariate logistic regression analysis and the LASSO algorithm to identify variables for constructing the clinical model. Finally, a combined model integrating radiomics features and clinical characteristics was developed. The dataset was partitioned into training (n = 100) and test (n = 26) sets at an 8:2 ratio. The predictive performance of the models was evaluated using ROC curves, while their clinical utility was assessed through decision curve analysis (DCA). RESULTS: Total of 1874 radiomics features were extracted from axial TWI-FS sequences. Following dimensionality reduction and feature selection, 14 radiomics features were identified as valuable. Among the radiomics models, the RF model demonstrated the highest predictive performance and generalizability, achieving AUC values of 0.796 in the training set and 0.740 in the test set. Subsequently, a clinical model was constructed using clinical characteristics, with the RF model exhibiting superior predictive performance and generalizability, yielding AUC of 0.876 (training set) and 0.817 (test set). Ultimately, the combined model integrating radiomics features and clinical characteristics demonstrated optimal predictive ability. The LR model achieved an AUC of 0.944 in the training set and 0.870 in the test set, while DCA confirmed its optimal clinical utility. CONCLUSION: The combined model integrating radiomics features and clinical characteristics demonstrated significant predictive performance and robustness in evaluating lesion necrosis extent following UAE for AM. Its discriminative capability surpassed that of single-modality prediction models, potentially offering a non-invasive objective assessment tool to optimize clinical decision-making pathways.