Enhanced CT radiomics-based features can effectively predict P504S expression status in renal cancer

ZHOU Jing; YANG Yuqiong; MA Yichuan; WANG Zhe; JI Ruoshi; WANG Ying; XU Jiali

doi:10.12122/j.issn.1674-4500.2025.07.08

Abstract：Objective To construct and validate a machine learning model based on enhanced CT radiomics features to predict the expression status of immunohistochemical index P504S in renal cancer.Methods Clinical,pathological and imaging data of 117 patients with pathologically confirmed renal carcinoma and defined P504S expression status in the First Affiliated Hospital of Bengbu Medical University from January 2019 to September 2024 were collected and retrospectively analyzed;Three-dimensional radiomics features from contrast-enhanced CT of the lesions were extracted to establish a predictive model for distinguishing between P504S-negative and P504S-positive cases.All cases were randomly divided into a training set and a test set at a ratio of 7:3.5-fold cross-validation was performed on the training set to select the optimal hyperparameters for establishing the predictive model,and the model using was validated the test set and the diagnostic performance of the model was analyzed using the ROC curve,calibration curve,and decision curve analysis.The region of interest was delineated based on the arterial and venous phases of CT scans.Data were normalized using Min-max normalization,and dimensionality reduction was performed through principal component analysis and Pearson similarity.The Relief algorithm was used for feature selection,and support vector machine and Naive Bayes were used as classifiers to construct the radiomics models for the arterial and venous phases,respectively.Results The radiomics model for the arterial phase achieved an AUC and accuracy of 0.801 and 0.805 on the training set and 0.833 and 0.743 on the test set,respectively.The radiomics model for the venous phase achieved an AUC and accuracy of 0.791 and 0.683 on the training set and 0.808 and 0.714 on the test set,respectively.The combined model for the arterial and venous phases achieved an AUC of 0.846 for all cases(95%CI:0.768-0.906),slightly higher than the radiomics model for the arterial phase(0.804,95%CI:0.720-0.871)and the radiomics model for the venous phase(0.823,95%CI:0.742-0.887),but there was no statistically significant difference(P>0.05).Conclusion A machine learning model based on enhanced CT radiomics features of renal cancer can predict the expression status of immunohistochemical indicator P504S.