Radiomics analysis of magnetic resonance imaging improves diagnostic performance of lymph node metastasis in patients with cervical cancer

Background and purpose: Robust parameters are needed to predict lymph node metastasis (LNM) in locally advanced cervical cancer patients in order to select optimal treatment regimen. The aim of this study is to utilize radiomics analysis of magnetic resonance imaging (MRI) to improve diagnostic performance of LNM in cervical cancer patients. Materials and methods: A total of 189 cervical cancer patients were divided into a training cohort (n = 126) and a validation cohort (n = 63). For each patient, we extracted radiomic features from intratumoral and peritumoral tissues on sagittal T2WI and axial apparent diffusion coefficient (ADC) maps. Afterward, the radiomic features associated with LNM status were selected by univariate ROC testing and logistic regression with the least absolute shrinkage and selection operator (LASSO) penalty in the training cohort. Based on the selected features, a support vector machine (SVM) model was established to predict LNM status. To further improve the diagnostic performance, a decision tree which combines the radiomics model with clinical factors was built. Results: Radiomics model of the intratumoral and peritumoral tissues on T2WI (T2_tumor+peri) showed best sensitivity and clinical LN (c-LN) status showed best specificity to predict LNM. The decision tree that combines radiomics model of T2_tumor+peri and c-LN status achieved best diagnostic performance, with AUC and sensitivity of 0.895 and 94.3%, 0.847 and 100% in the training and validation cohort respectively. Conclusions: The decision tree, which incorporates radiomics model of T2_tumor+peri and c-LN status can be potentially applied in the preoperative prediction of LNM in locally advanced cervical cancer patients.