Preoperative Prediction of Adherent Perirenal Fat.

Importance: For patients with kidney tumors requiring surgical treatment, adhesive perirenal fat is a frustrating variable that surgeons encounter during surgery, but the current image-dependent kidney morphometric scoring system used to predict the potential difficulty of surgery ignores this factor. Accurate preoperative prediction of perirenal fat status remains an urgent need. Purpose: To determine whether radiomics features of perirenal fat derived from computed tomography images can provide valuable information for judging perirenal fat status, develop a prediction model based on CT radiomics combined with deep learning, and validate the performance of the model in an independent cohort. Design, setup and participants: The study included one retrospective dataset and one prospective dataset from four medical centers between January 2020 and September 2023. Kidney plain CT scan was performed in xx adult patients with partial nephrectomy or radical nephrectomy. The training set, validation set, and internal test set were provided by the First Hospital of Jilin University, and the external test set was provided by the First Hospital of Siping City, Liaoyuan Central Hospital and Dongfeng County Hospital. This diagnostic study used single-institution data from January 2020 to May 2023 to extract imaging omics features from the perirenal fat region (independent sample T-test, minimum absolute contraction, and selection operator logistic regression was used to screen for the best imaging omics features). Univariate and multivariate analyses of clinical variables in patients prior to renal surgery were performed to determine independent predictors of adherent perirenal fat in the clinical setting. Different classifiers were used to build prediction models using only the image-omics features and fusion prediction models using independent clinical predictors combined with the image-omics features. Its performance is verified in two test sets. Main achievements and measures: The discriminant performance of the image omics model was evaluated by the area under the receiver operating characteristic curve and confirmed by decision curve analysis.