Related Experiment Video
Updated: Apr 24, 2026
Spatial Measurements of Perfusion, Interstitial Fluid Pressure and Liposomes Accumulation in Solid Tumors
Published on: August 18, 2016
Preoperative contrast-enhanced CT-based radiomics model for distinguishing retroperitoneal well-differentiated
Peidang Fan1, Jiulong Zhang2, Jiongyuan Wang3
1Department of General Surgery, Zhongshan Hospital, Fudan University, Shanghai, China; Department of General Surgery, Shanghai Xuhui Center Hospital, Shanghai, China.
Background:
This retrospective multi-center study investigated whether radiomics features extracted from contrast-enhanced computed tomography (CECT) images could distinguish between the two most common pathological subtypes of retroperitoneal sarcoma (RPS): dedifferentiated liposarcoma (DDLPS) and well-differentiated liposarcoma (WDLPS), as well as differentiate histological grading in DDLPS.
Materials And Methods:
Patients with localized RPS who underwent surgery with curative intent between 2009 and 2021 were identified, along with their preoperative CECT images, which were obtained from two tertiary hospitals. Volumes of interest (VOIs) were constructed by segmenting tumor regions on CT images to extract radiomics features. In the training set, univariate logistic regression analysis and the least absolute shrinkage and selection operator (LASSO) algorithm were employed to identify the optimal radiomics features and construct the models. These models were then validated using internal and external validation sets. The models' utility was evaluated using the area under the receiver operating characteristic (ROC) curve, accuracy, the calibration curve, and decision curve analysis (DCA).
Results:
Compared with multi-phase radiomics model, venous-phase radiomics model exhibits comparable predictive performance in differentiating WDLPS and DDLPS (AUCs: 0.848-0.852), as well as sclerosing WDLPS and DDLPS (AUCs: 0.811-0.838), with higher diagnostic accuracy than radiologists. In cases with discordant biopsy and final pathology (n = 5), the model demonstrated 80% accuracy. The analysis of calibration curves and DCA shows that the model is well calibrated and has significant clinical advantages. However, the venous-phase radiomics model shows only moderate predictive performance in distinguishing DDLPS grade 2 and 3.
Conclusion:
The venous-phase radiomics model has strong predictive ability in preoperative differentiation between WDLPS and DDLPS, as well as between sclerosing WDLPS and DDLPS. This makes it a promising imaging biomarker that could facilitate personalized management and precision medicine.
