# Study on the prediction of postoperative metastasis in renal cancer using perirenal fat CT radiomics combined with clinical features

## TL;DR

This study creates a model combining CT scans and clinical data to predict if kidney cancer will spread after surgery.

## Contribution

A novel predictive model integrating clinical data, tumor radiomics, and perirenal fat radiomics for metastasis prediction in RCC.

## Key findings

- The combined model achieved an AUC of 0.958, outperforming individual models.
- PRF radiomics and tumor radiomics models showed strong predictive performance (AUCs of 0.841 and 0.848).
- Clinical factors like tumor diameter and neutrophil count were significant predictors.

## Abstract

This study aims to develop a combined predictive model for predicting postoperative metastasis risk in renal cell carcinoma (RCC) patients based on preoperative arterial-phase Computed tomography(CT) images, integrating clinical data, perirenal fat (PRF), and tumor radiomics features.

A retrospective analysis was conducted on abdominal CT images and clinical data of patients with pathologically confirmed renal cell carcinoma. Inclusion criteria included preoperative CT scanning, biopsy or surgical confirmation of RCC, and postoperative follow-up to assess metastasis status. Exclusion criteria included patients who had undergone endocrine or anti-tumor treatments. The TotalSegmentator model was used for bilateral PRF segmentation, and radiomics features were extracted. Clinical models, PRF radiomics models, and tumor radiomics models were constructed and integrated into a combined predictive model (Nomogram). The performance of the models was evaluated using receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) values.

A total of 120 patients were included, with 36 (30%) developing postoperative metastasis. The clinical model (AUC = 0.877) identified tumor maximum diameter and neutrophil count as independent predictive factors. The PRF radiomics model (AUC = 0.841) and tumor radiomics model (AUC = 0.848) performed well. The combined model (Nomogram) achieved an AUC of 0.958, significantly outperforming the individual models. All models demonstrated good calibration, and decision curve analysis confirmed their clinical net benefit.

The combined predictive model developed in this study, integrating preoperative clinical data, PRF, and tumor radiomics features, effectively predicts postoperative metastasis risk in RCC patients. This model provides valuable non-invasive information for preoperative metastasis risk assessment and offers reliable guidance for personalized treatment plans, highlighting the critical role of the tumor microenvironment in RCC progression.