To explore the value of conventional MRI radiomics for predicting occult cervical lymph nodes (LNs) metastases in early-stage oral tongue squamous cell carcinoma (OTSCC).

The preoperative MRI data of 77 cases of early-stage OTSCCs (cT1-2N0M0) in Shanghai Ninth People′s Hospital from January 2015 to December 2019 were retrospectively analyzed. All patients underwent primary lesion resection with selective neck dissection and the pathologic reports of LNs couldal be obtained. In total, 168 LNs (51 positive and 117 negative metastases) were enrolled and allocated into training set (n=112) and validation set (n=56) with a ratio of 2∶1 using random number table. The volumes of interest of LNs on T₂WI and contrast enhanced T₁WI (ceT₁WI) were delineated by two doctors using ITK-SNAP software. The 1 046 radiomics features of each sequence were extracted using 3D Slicer software. Data dimension reduction was done by inter-observer agreement analysis and univariate analysis. The least absolute shrinkage and selection operator regression analysis were used for selecting optimal feature subsets and constructing radiomics signature for each sequence. Mann-Whitney U test was used to compare the differences of node size and radiomics scores between the LNs with positive and negative metastases. The receiver operating characteristic (ROC) curve was used to explore the performance of LNs size, T₂WI radiomics signature and ceT₁WI radiomics signature in predicting occult LNs metastases. Stepwise logistic regression was used to determine the independent predictors.

Fifteen and 10 optimal features were selected to construct radiomics signature for T₂WI and ceT₁WI respectively. The short diameter, T₂WI radiomics signature and ceT₁WI radiomics signature showed significant differences between LNs with positive and negative metastases in the both training and validation sets (all P<0.05), with the areas under the ROC curve of 0.67, 0.83 and 0.82 in the training set, and 0.69, 0.78 and 0.70 in the validation set, respectively. In the stepwise logistic regression analysis, T₂WI radiomics signature was identified as the independent predictor in the both sets (training set: OR=5.92, P<0.001; validation set: OR=2.53, P=0.012).

Conventional MRI radiomics can provide a good potential to predict occult LNs metastases in early-stage OTSCC.

Carcinoma, squamous cell; Tongue neoplasms; Magnetic resonance imaging; Lymph nodes; Metastases; Radiomics