Prognostic Prediction of NPC Based on MR Diffusion-weighted Imaging

Prognostic Prediction of Nasopharyngeal Carcinoma Based on Radiomics Features of MR Diffusion-weighted Imaging

The purpose of this study is to explore whether the imaging model based on RESOLVE-DWI sequence can exploiting the heterogeneity of nasopharyngeal carcinoma and indicate the prognosis, so as to provide intervention information for clinical decision-making. All patients were randomly divided into the training group and the validation group. Radiomics features extracted from T2-weighted, DWI, apparent diffusion coefficient (ADC), and contrast- enhanced T1-weighted were used to build a radiomics model. Patients'clinical variables were also obtained to build a clinical model. Model of training cohort was established using cross-validation for nasopharyngeal carcinoma prognosis by machine learning, including Logistics Regression, SVM, KNN, Decision Tree, Random Forest, XGBoost, and then, the model will be verified in the validation cohort. Area under the curve (AUC) of the Machine learning model was used as the main evaluation metric.

Inclusion Criteria:

1. patients with nasopharyngeal carcinoma diagnosed by pathology;
2. complete clinical data and MR imaging data;
3. without radiotherapy, chemotherapy or operation before MR examination.

Exclusion Criteria:

1. incomplete follow-up data;
2. poor image quality and can not be used for analysis;
3. patients with other tumors in the past or at the same time.