Application Of Nomogram Model Based On Multi-Modal MRI Radiomics And Clinical Features To Predict Cervical Lymph Node Metastasis In Patients With Tongue And Oral Floor Squamous Cell Carcinoma

Objective: Accurate preoperative evaluation of cervical lymph node status in patients with tongue and oral floor squamous cell carcinoma is significant for the selection of treatment strategies.The purpose of this study was to develop and validate a nomogram model based on multi-modal MRI radiomics and clinical features for the preoperative prediction of the risk of cervical lymph node metastasis in patients with tongue and oral floor squamous cell carcinoma.Methods: The study retrospectively analyzed clinical data and MRI data of 400 patients with tongue and oral floor squamous cell carcinoma diagnosed by pathology.All 400 cases included T1 WI and FS-T2 WI to form a two-sequence MRI group,which was randomly divided into training set and validation set.195 cases included T1 WI,FS-T2 WI and DCE-MRI to form a three-sequence MRI group,which was randomly divided into training set and validation set.After segmenting the whole tumor in each MRI sequence,1130 radiomics features were extracted from T1 WI,FS-T2 WI and DCE-MRI,respectively.After screening and dimensionality reduction of radiomics features in training set,the random forest algorithm was used to construct the radiomics model.Significant clinical features were screened in the training set and clinic models were constructed.Calculate the Radscore of the radiomics model and integrate the Radscore with the clinical features to construct the combined model.The three models were validated in the c N0 subgroup patients.Subgroup analysis was performed according to the clinical T stage.The receiver operation characteristic curve was drawn,and the AUC,accuracy,sensitivity and specificity of each model in the training set and validation set were calculated respectively to evaluate the predictive performance of each model.Results: With the increase of sequences,the performance of the radiomics model and combined model is improved.In the two-sequence MRI group,the AUC values of the combined model in training set and validation set were 0.926(0.893-0.955)and0.841(0.766-0.903).In the three-sequence MRI group,the AUC values of the combined model were improved,and the AUC values of training set and validation set were 0.962(0.929-0.984)and 0.899(0.813-0.973).The combined model could identify lymph node metastasis in the c N0 subgroup.In the c N0 subgroup analysis,the AUC value of the combined model of the two-sequence MRI group in training set and validation set were 0.847(0.770-0.910)and 0.757(0.612-0.894),and the AUC values of the combined model of the three-sequence MRI group in training set and validation set were 0.946(0.884-0.992)and 0.808(0.610-0.967).With the increase of tumor stage,the performance of the combined model improved.In the two-sequence MRI group,the AUC values of the combined model in the c T1-T2 subgroup were 0.895(0.833-0.941)and 0.826(0.726-0.920)in training set and validation set,and the AUC values of the combined model in the c T3-T4 subgroup were 0.940(0.903-0.975)and0.860(0.759-0.944)in training set and validation set.In the three-sequence MRI group,the AUC values of the combined model in the c T1-T2 subgroup were0.935(0.867-0.986)and 0.841(0.648-0.982)in training set and validation set,and the AUC values of the combined model in the c T3-T4 subgroup were 0.970(0.928-0.998)and 0.967(0.889-1.000)in training set and validation set.Conclusion: The nomogram model based on multi-modal MRI radiomics and clinical features can predict the risk of cervical lymph node metastasis in patients with tongue and oral floor squamous cell carcinoma before surgery,which is helpful to provide a basis for accurately evaluating the status of cervical lymph node and assist in selecting scientific treatment strategies.