基于MRI影像组学、临床及超声特征术前预测子宫内膜癌淋巴结转移：双中心临床研究

PURPOSE: To explore MRI-based radiomic, clinical, and sonographic features for constructing models to predict lymph node metastasis (LNM) in endometrial carcinoma (EC) preoperatively and noninvasively.

METHODS: From January 2016 to December 2024, clinical, imaging and pathological information of patients with EC were retrospectively collected at center 1 and 2. Radiomic features extracted from preoperative MRI were screened, a radiomic model was developed and radscore was calculated. Univariate and multivariate analyses were performed with all features to identify independent LNM predictors. Another three models were constructed: a clinical model based on clinical features, an ultrasound model based on features defined by International Endometrial Tumor Analysis, and a combined model integrating all predictors. Receiver operator characteristic curve and DeLong test were employed for evaluating model performance. Clinical decision curve (CDC) and calibration curve were used for assessing the clinical benefit of the models.

RESULTS: 225 patients with EC were included, in which 44 patients were with LNM and 181 patients were without LNM. Statistical analyses showed that radscore, age, CA125, and endometrium thickness (EMT) were predictors for LNM. DeLong test demonstrated that CombinedModel had the superior predictive performance, which was presented as a nomogram, with area under the curve values of 0.895 (95%CI:0.828,0.961),0.870 (95%CI:0.739,1.000), 0.890 (95%CI: 0.766,1.000) for training, internal and external validation set, respectively. CDC and calibration curve of the nomogram demonstrated higher clinical benefit than other models.

CONCLUSION: The nomogram comprising age, CA125, radscore, and EMT exhibited the best performance in LNM prediction of patients with EC, making it a valuable tool for decision-making.