A Nomogram for Preoperative Prediction of Tumor Aggressiveness and Lymphovascular Space Involvement in Patients with Endometrial Cancer

Abstract

Background/Objectives: To develop a nomogram for predicting tumor aggressiveness and the presence of lymphovascular space involvement (LVSI) in patients with endometrial cancer (EC) using preoperative MRI and pathology-laboratory data. Methods: This IRB-approved, retrospective, multicenter study included 245 patients with histologically confirmed EC who underwent preoperative MRI and surgery at participating institutions between January 2020 and December 2024. Tumor type and grade, both from preoperative biopsy and surgical specimens, as well as preoperative CA125 and HE4 levels, were retrieved from institutional databases. A preoperative MRI was used to assess tumor morphology (polypoid vs. infiltrative), maximum diameter, presence and depth (< or >50%) of myometrial invasion, cervical stromal invasion (yes/no), and minimal tumor-to-serosa distance. The EC-to-uterus volume ratio was also calculated. Results: Among the 245 patients, 27% demonstrated substantial LVSI, and 35% were classified as aggressive on final histopathology. Multivariate analysis identified independent MRI predictors of LVSI, including cervical stromal invasion (OR = 9.06; p = 0.0002), tumor infiltration depth (OR = 2.09; p = 0.0391), and minimal tumor-to-serosa distance (OR = 0.81; p = 0.0028). The LVSI prediction model yielded an AUC of 0.834, with an overall accuracy of 78.4%, specificity of 92.2%, and sensitivity of 43.1%. For tumor aggressiveness prediction, significant predictors included biopsy grade (OR = 8.92; p < 0.0001), histological subtype (OR = 12.02; p = 0.0021), and MRI-detected serosal involvement (OR = 14.39; p = 0.0268). This model achieved an AUC of 0.932, with an accuracy of 87.0%, sensitivity of 79.8%, and specificity of 91.2%. Both models showed excellent calibration (Hosmer-Lemeshow p > 0.86). Conclusions: The integration of MRI-derived morphological and quantitative features with clinical and histopathological data allows for effective preoperative risk stratification in endometrial cancer. The two nomograms developed for predicting LVSI and tumor aggressiveness demonstrated high diagnostic performance and may support individualized surgical planning and decision-making regarding adjuvant therapy. These models are practical, reproducible, and easily applicable in standard clinical settings without the need for radiomics software, representing a step toward more personalized gynecologic oncology.

Keywords: endometrial neoplasms; lymphovascular space invasion; magnetic resonance imaging; neoplasm grading; nomogram model; uterus.

Figure 1

Nomogram for predicting the probability of lymphovascular space invasion (LVSI) on definitive pathological specimen in patients with endometrial cancer, based on logistic regression analysis. To use the nomogram, locate the patient-specific value for each variable on its corresponding axis, draw a vertical line up to the “Points” axis to determine the individual score, then sum all the points and find the total on the “Total Points” axis. Finally, draw a vertical line downward to determine the predicted probability at the bottom of the chart. This graphical tool enables intuitive, individualized risk assessment using routinely available clinical and imaging data. The contribution of each variable is weighted according to its odds ratio in the logistic regression model.

Figure 2

Nomogram for predicting the probability of tumor aggressiveness in endometrial cancer based on the logistic regression model. To use the nomogram, locate the patient-specific value for each variable on its corresponding axis, draw a vertical line up to the “Points” axis to determine the individual score, then sum all the points and find the total on the “Total Points” axis. Finally, draw a vertical line downward to determine the predicted probability at the bottom of the chart. This graphical tool enables intuitive, individualized risk assessment using routinely available clinical and imaging data. The contribution of each variable is weighted according to its odds ratio in the logistic regression model.