Noninvasive MRI-based radiomics and clinical features nomogram for preoperative prediction of HER-2 expression in bladder urothelial carcinoma

HER-2 is an emerging therapeutic target in bladder urothelial carcinoma (UC). Conventional assessment by biopsy is invasive and prone to sampling bias. We aimed to develop a noninvasive MRI-based radiomics nomogram integrating clinical predictors for preoperative HER-2 expression. Ninety-one UC patients who underwent multiparametric MRI were retrospectively analyzed. Radiomics features were extracted, filtered by LASSO regression, and used to build machine learning models. Independent clinical predictors were identified by logistic regression. A combined nomogram was constructed and evaluated using AUC, calibration, and decision curve analysis. Age and tumor grade were independent clinical predictors of HER-2 expression. The radiomics model achieved strong discrimination (AUC = 0.871 training; 0.842 validation). The clinical model showed moderate performance. The integrated nomogram outperformed both, with AUC = 0.901 (95% CI 0.832–0.970) in the training cohort and 0.956 (95% CI 0.875–1.000) in the independent test cohort, showing improved discriminative performance, with calibration and clinical utility that should be interpreted cautiously. The combined MRI-based radiomics-clinical nomogram showed promising performance for predicting HER-2 overexpression in bladder urothelial carcinoma and may serve as a complementary tool for noninvasive preoperative risk stratification. However, further validation in larger prospective multicenter cohorts is required before clinical implementation.