基于多种机器学习模型预测膀胱癌病理分级的价值研究OA

Objective:To establish a predictive model for bladder cancer pathological grading using multi-phase CT ra-diomics and integrated clinical indicators.Methods:A retrospective analysis was conducted on CT and clinical data from 208 patients diagnosed with bladder cancer.The data were divided into a training group(n=166)and a validation group(n=42)in an 8∶2 ratio.The Radcloud research platform was used to perform layer-by-layer segmentation of plain,arterial,and venous phase images to obtain regions of interest and extract radiomics features.Redundant features were excluded using variance se-lection,univariate selection,and LASSO algorithms.Three machine learning algorithms were used to construct nine models based on plain scans,enhanced scans,and combined scans.The model with the best predictive performance was selected and combined with clinical independent risk factors to establish a fusion model.The predictive performance of each model was e-valuated using receiver operating characteristic(ROC)curves.Results:Models established based on different phases and classi-fiers demonstrated good predictive performance.Among them,the radiomics model constructed using the BernoulliNB classifier based on combined scans exhibited the highest comprehensive diagnostic performance,with an AUC of 0.836 in the training group and 0.820 in the validation group.The fusion model,established by combining it with independent risk factors,showed that the predictive model using the BernoulliNB classifier had the best performance,with an AUC of 0.781 in the training group and 0.861 in the validation group.Conclusion:The fusion model established based on multi-phase CT radiomics features and clinical independent risk factors demonstrated good predictive performance in the pathological grading of bladder cancer.