基于生境放射组学-临床模型预测膀胱癌术后早期复发

PURPOSE: A high recurrence rate is a serious problem in the bladder cancer (BC) treatment. To develop a habitat-based radiomics-clinical model for noninvasive recurrence prediction in BC patients undergoing transurethral resection of bladder tumor (TURBT) or radical cystectomy (RC).

METHODS: We retrospectively enrolled 294 BC patients who underwent TURBT or RC at our Hospital. Tumor regions of interest (ROIs) were automatically segmented and manually checked. Voxel-wise radiomic features were extracted. The K-means clustering algorithm was applied to perform cluster analysis on the extracted features. The Calinski-Harabasz (CH) index was calculated to determine the optimal number of clusters (the one yielding the highest CH value). Subsequently, features from these subregions were extracted and further filtered using the t-test and Pearson correlation analysis. The final feature set was selected through the least absolute shrinkage and selection operator regression. The dataset was randomly divided into the training (n = 235) and validation (n = 59) sets (8:2 ratio) for model training and evaluation. The habitat-based radiomics model based on Support Vector Machine (SVM) was developed based on the final feature set. Additionally, we combined clinical risk factors to establish and validate a radiomics-clinical model.

RESULTS: We divided the tumor ROIs into three subregions. The Areas Under the Curve (AUCs) for the habitat-based radiomics-model were 0.870 (95% confidence interval [CI]: 0.825-0.914) and 0.865 (95% CI: 0.774-0.956) in the training and validation set. The radiomics-clinical model, incorporating tumor grades and intravesical therapy, achieved AUCs of 0.883 (95% CI: 0.842-0.924) and 0.956 (95% CI: 0.912-1.000) in the training and validation sets, respectively.

CONCLUSIONS: The habitat-based radiomics-clinical model demonstrated superior performance in predicting BC recurrence.