CT-based radiomics nomogram for differentiating dedifferentiated liposarcoma from well-differentiated liposarcoma.

Purpose: This study aimed to use radiomics features derived from plain CT scans to construct a model that can predict the pathological classification of Retroperitoneal liposarcoma (RLPS) preoperatively, help enhance preoperative planning and inform tailored treatment strategies. Methods: This retrospective study involving 114 consecutive RLPS patients from January 2022 to December 2024. Clinical, pathological, and CT imaging data were gathered. Radiomics features were extracted from plain CT scans and were selected through Least Absolute Shrinkage and Selection Operator (LASSO) regression. A radiomics signature was created, and a nomogram was developed for predicting dedifferentiated liposarcoma (DDLPS). Performance of the nomogram was assessed and compared with radiologist evaluation of the CT imaging. Area under the curve (AUC) and decision curve analysis in both training and validation sets. Results: Higher Ki-67 and unclear tumor boundary was established as an independent predictor for DDLPS. Five radiomics features were identified as significant predictors. a nomogram was developed by combining these features. The nomogram showed an AUC of 0.91 (95% CI: 0.84-0.98) and 0.89 (95% CI: 0.73-1.00) in the training and validation set, which outperforming the radiologist evaluation. Decision curve analysis confirmed that the nomogram provided a higher net clinical benefit compared to the radiologist. Conclusions: The radiomics nomogram significantly enhances the preoperative differentiation of RLPS subtypes.