Abstract
BACKGROUND: This study aims to develop a habitat model based on vaginal ultrasound images for the non-invasive prediction of LVSI in endometrial cancer (EC) patients. METHODS: We retrospectively collected data from 871 EC patients who underwent surgical resection from September 2017 to January 2025, dividing them into LVSI and non-LVSI groups. Patients’ age and gender were matched using Propensity Score Matching (PSM) in a 1:1 ratio to control for potential confounding factors. Selected patients were randomly divided into training and testing groups in a 7:3 ratio. We extracted habitat features and radiomic features from transvaginal ultrasound images, established a combined machine learning (ML) model after dimensionality reduction, and evaluated the proposed model’s utility using receiver operating characteristic and decision curve analysis. RESULTS: A total of 250 EC patients were included in the study. In the training group, the AUC for the radiomic model was 0.809 (0.745–0.873),the AUC for the habitat model was 0.862 (0.809–0.916),and the AUC for the combined model was 0.874 (0.824–0.924).In the testing group, the AUC for the radiomic model was 0.779(0.667–0.892), the AUC for the habitat model was 0.794(0.692–0.895),and the AUC for the combined model was 0.849(0.758–0.939).The combined model outperformed the single radiomic and habitat models. Decision curve analysis confirmed the clinical utility of the combined model. CONCLUSION: The habitat model based on ultrasound images can accurately and non-invasively distinguish LVSI in EC patients, aiding doctors in developing more favorable treatment plans for patients.