2345 - Construction of a Lymph Node Metastasis Prediction Model for Endometrial Cancer Based on MRI Radiomics

Z. Li¹, Y. Su², G. Yu³, Y. Yin⁴, and Z. Li⁵; ¹Shandong Province Key Laboratory of Medical Physics and Image Processing Technology, School of Physics and Electronics, Shandong Normal University, jinan, shandong, China, ²Yantai Yuhuangding Hospital, yantai, shandong, China, ³Shandong Province Key Laboratory of Medical Physics and Image Processing Technology, School of Physics and Electronics, Shandong Normal University, Jinan, China, ⁴Department of Radiation Physics, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China, ⁵Department of Radiation Oncology Physics and Technology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China

Purpose/Objective(s): To evaluate the performance of multisequence magnetic resonance imaging (MRI)-base radiomics features combined with clinical parameters in predicting lymph node metastasis (LNM) in patients with type I endometrial cancer (EC) to assist in preoperative lymphadenectomy decisions. Materials/

Methods: A total of 167 patients with type I EC were included (118 in the training set and 49 in the test set). Radiomics features were extracted from dynamic contrast-enhanced T1-weighted images (T1), T2-weighted images (T2), and apparent diffusion coefficients (ADC) by AccuContour V3.0 software. Spearman and the least absolute shrinkage and selection operator (LASSO) were used as features selection methods. We also developed radiomics models (combining three MRI sequences) and clinical models using three classifiers, logistic regression (LR), support vector machine (SVM) and random forest (RF). Finally, clinical+radiomics models were created based on the radiomics models and the clinical models. The models were evaluated using the area under the curve (AUC) of receiver operating characteristic and decision curve analysis (DCA).
Results: Among the 1409 features, Spearman and LASSO finally selected 12 radiomics features. The RF classifier had the best performance. The AUC of the radiomics, clinical and clinical+radiomics model based on the RF classifier in the test set was 0.926, 0. 679 and 0.978, the sensitivity was 1.000, 0.364 and 1.000, the specificity was 0.921, 0.711 and 0.974, the accuracy was 0.939, 0.633 and 0.980, the positive predictive value (PPV) was 0.786, 0.267 and 0.917, the negative predictive value (NPV) was 1.000, 0.794 and 1.000, respectively. DCA demonstrated that the RF clinical+radiomics model was predictive of LNM and yielded net clinical benefit.
Conclusion: This study proposed a RF clinical+radiomics Nomogram that combined multisequence radiomics features with clinical parameters. It could improve the accuracy of preoperative prediction of LNM in type I EC and provide an aid in making decisions about lymphadenectomy.