Work

Multiplanar MRI-Based Predictive Model for Preoperative Assessment of Lymph Node Metastasis in Endometrial Cancer

Year: 2019

Type: article

Abstract: Introduction: Assessment of lymph node metastasis (LNM) is crucial for treatment decision and prognosis prediction for endometrial cancer (EC). However, the sensitivity of the routinely used magnetic resonance imaging (MRI) is low in assessing normal-sized LNM (diameter 0cm~0.8cm).We aimed to develop a predictive model based on magnetic resonance (MR) images and clinical parameters to predict LNM in normal-sized lymph nodes (LNs). Materials and Methods: A total of 200 retrospectively patients were enrolled and divided into a training cohort (n=140) and a test cohort (n=60). All patients underwent preoperative MRI and had pathological result of LNM status. In total 4179 radiomic features were extracted. Four models including a clinical model, a radiomic model and two combined models were built. Area under the receiver operating characteristic (ROC) curves (AUC) and calibration curves were used to assess these models. Subgroup analysis was performed according to LN size. All patients underwent surgical staging and had pathological results. Results: All of the four models showed predictive ability in LNM. One of the combined models, ModelCR1, consisting of radiomic features, LN size, and cancer antigen 125, showed the best discrimination ability on the training cohort (AUC, 0.892; 95% confidence interval [CI], 0.834-0.951) and test cohort (AUC, 0.883; 95%CI, 0.786-0.980). The subgroup analysis showed that this model also indicated good predictive ability in normal-sized LNs (0.3cm~0.8cm group, accuracy = 0.846; <0.3cm group, accuracy = 0.849). Furthermore, compared with the routinely preoperative MR-report, the sensitivity and accuracy of this model had a great improvement. Conclusions: A predictive model was proposed based on MR radiomic features and clinical parameters for LNM in EC. The model had a good discrimination ability, especially for normal-sized LNs.

Source: Frontiers in Oncology

DOI: 10.3389/fonc.2019.01007