‌CT image-derived radiomics predicts molecular subtypes in bladder urothelial carcinoma: validation of a non-invasive classification strategy.

This pilot study aimed to investigate the correlation between CT-based radiomic features and molecular subtypes in bladder urothelial carcinoma, to determine whether pretreatment computed tomography (CT)-derived radiomic profiles can discriminate distinct molecular classifications of bladder cancer. We retrospectively analyzed 96 patients with pathologically confirmed bladder urothelial carcinoma who underwent transurethral resection of bladder tumor (TURBT). Radiomic texture parameters including mean intensity, standard deviation, entropy, kurtosis, and skewness were extracted from preoperative CT images. Statistical analyses using SPSS 26.0 evaluated associations between these parameters and molecular subtypes (basal vs. luminal), with statistical significance defined as P < 0.05. The basal subtype demonstrated significantly higher mean intensity (P = 0.016) and entropy values (P < 0.001) compared to the luminal subtype. Receiver operating characteristic (ROC) analysis identified entropy as the most robust predictor of molecular classification, achieving an area under the curve (AUC) of 0.790 (95% CI: 0.685-0.895) with an optimal cutoff value of 4.733. CT-based radiomic texture analysis shows potential for non-invasive discrimination of molecular subtypes in bladder urothelial carcinoma, with entropy exhibiting superior diagnostic performance in molecular classification prediction.