Differential Diagnosis Analysis Of Nasopharyngeal Carcinoma And Non-nasopharyngeal Carcinoma By MR Radiomics And Machine Learning Of Metastatic Cervical Lymph Nodes

Objective: To analyze the radiomics features of metastatic cervical lymph nodes from nasopharyngeal carcinoma and non-nasopharyngeal carcinoma using magnetic resonance imaging,and to identify the best radiomics model to distinguish nasopharyngeal carcinoma and non-nasopharyngeal carcinoma.Methods: The preoperative cervical MR data and clinical data of patients with head and neck tumors with cervical lymph node metastasis confirmed by pathology were collected retrospectively,and they were divided into two groups: source of nasopharyngeal carcinoma and non-nasopharyngeal carcinoma.The region of interest was drawn in the MRI-T1 WI enhanced sequence and radiomics features were extracted,the optimal features were screened,and seven machine learning models were selected for classification training to explore the differential diagnosis efficacy of radiomics in nasopharyngeal carcinoma and non-nasopharyngeal carcinoma.Results: dimensionality reduction feature screening,eight optimal features were finally selected to enhance the establishment of the T1 WI radiomics classification model.Among the four machine learning models,Among the four machine learning models,the radiomics model established by the Logistic Regression(LR)method showed better diagnostic performance,The accuracy of the test set reached 0.875,The area under the receiver operating curve was 0.865,and the sensitivity and specificity were 0.786 and 0.944,respectively.Conclusion: The differential model based on MRI imaging of metastatic cervical lymph nodes can noninvasively distinguish nasopharyngeal carcinoma from non-nasopharyngeal carcinoma.