Clinical Application Study Of Ultrasound Image-based Radiomics In Predicting Pathological Grading Of Bladder Cancer

Objectives: This study aimed to establish a predictive model for pathological grading of bladder cancer based on ultrasound image-based radiomics using machine learning algorithms,with a focus on preoperatively predicting the pathological grading of non-invasive bladder tumors.Methods: A total of 180 bladder cancer patients were recruited(median age67 years,range 39-91 years;142 males,38 females)and randomly divided into a training set(n=126)and a validation set(n=54)in a 7:3 ratio.Ultrasound images were acquired,and tumor contours were manually delineated and underwent image segmentation and preprocessing.A total of 835 features were extracted and then subjected to dimensionality reduction and feature selection using the minimum redundancy maximum relevance algorithm(m RMR)and the least absolute shrinkage and selection operator algorithm(LASSO)in the training set to construct a radiomics score(Rad-Score).The resulting meaningful feature variables were then used in several machine learning algorithms to establish a predictive model for preoperative prediction of pathological grading of non-invasive bladder cancers.Results:(1)There was no statistically significant difference in age distribution,gender ratio,or pathological grading ratio between the training and validation sets.(2)Thirteen selected features were used to calculate the Rad-Score,which was significantly higher for high pathological grading bladder tumors(0.67[0.43-0.85])than for low pathological grading bladder tumors(0.33 [0.15-0.43])in the training set(P<0.005).Similarly,in the validation set,the Rad-Score was also significantly higher for high pathological grading bladder tumors(0.63[0.44-0.91])than for low pathological grading bladder tumors(0.35 [0.13-0.49])(P<0.005).In addition,the calibration curve demonstrated good discrimination and accuracy for the combination of the Rad-Score with the clinical features.Conclusion : Ultrasound image-based radiomics using machine learning algorithms can precisely predict the pathological grading of non-invasive bladder cancers preoperatively,which has clinical significance for personalized treatment of bladder cancer.