Objective Preoperative non-invasive prediction of microsatellite instability (MSI) in endometrial carcinoma (EC) based on ultrasound radiomic features and clinical information provides a reference for clinical prevention, treatment, and personalized therapy.

Methods The clinical and ultrasonographic data of 60 patients with EC confirmed by surgical pathology at the Tianjin Medical University Cancer Institute & Hospital from June 2022 to June 2024 were collected for retrospective analyses. The patients were assigned into MSI (25 patients) and microsatellite stable (MSS) (35 patients) groups. Radiomic features were extracted from ultrasound images in the following four distinct regions of interest (ROIs): intratumor region, intratumor + 1 mm peritumoral region, intratumor + 2 mm peritumoral region, and intratumor + 3 mm peritumoral region. Radiomics models were constructed based on the features of each ROI. The radiomics model that demonstrated the highest predictive performance was selected and integrated with clinical factors to build a comprehensive model, and the predictive efficacy of each model was assessed using receiver operating characteristic (ROC) curve analysis.

Results Significant differences in patient age, endometrial thickness, and personal history of previous malignancies (all P < 0.05) were observed between MSI and MSS groups. The intratumor + 2 mm peritumoral radiomics model achieved the highest AUC (0.80) among the four radiomics models. The clinical model achieved an AUC of 0.88, whereas the comprehensive model achieved the highest AUC overall (0.97). The comprehensive model significantly outperformed both the clinical and best-performing radiomics models in sensitivity, specificity, and accuracy (P < 0.05).

Conclusions Thecomprehensive model constructed by combining clinical information with ultrasound radiomics features of intratumor + peritumor imaging provides high clinical value in preoperatively predicting the MSI of EC, which can aid in creating a personalized treatment plan for each patient.