A Prediction Model for Para-Aortic Lymph Node Metastasis in Cervical Cancer by Radiomics Analysis Using Pre-Treatment MRI Images of the Primary Tumor

Whole pelvis irradiation (WPI) is the standard radiation technique for locally advanced cervical cancer without para-aortic lymph node (PALN) metastasis. PALN is one of the most common late failure sites in patients treated with WPI. Although there are several reports about the utility of prophylactic extended-field irradiation (EFI), it is still controversial which patients benefit from prophylactic EFI. If we could predict a PALN metastasis from pre-treatment imaging data, it might help patients to select prophylactic EFI. This study aimed to construct a predictive model for the PALN metastasis in patients with cervical cancer by radiomics analysis using pretreatment magnetic resonance imaging (MRI) of the primary tumor.Data of 94 patients with cervical squamous cell carcinoma who underwent radiotherapy between 2003/10 and 2018/2 were split into two sets: 66 patients for the training/validation and 28 patients for testing. The PALN status was classified into two groups (positive or negative). Both the synchronous and metachronous PALN metastasis was classified as PALN positive. A total number of 9394 radiomics features per a patient image were extracted from T1- and T2-weighted MRI images. The set of candidate predictors were selected with the least absolute shrinkage and selection operator (LASSO) logistic regression and build predictive models with neural network classifiers were used. The precision, accuracy, and sensitivity by generating confusion matrices and the areas under the receiver operating characteristic curve (AUC) for each model were evaluated.By the LASSO analysis of the training/validation data, we found 9 radiomics features from T1-weighted MRI image and 61 radiomics features from T2-weighted MRI image for the classification. The accuracy, specificity, sensitivity, and AUC of the prediction model for the dataset in testing group were 67.9 %, 91.0%, 10.0%, and 0.60 with T1-weighted MRI image, 95.7%, 97.0%, 92.5%, and 0.98 with T2-weighted MRI image, 96.4%, 99.0%, 90.7%, and 0.99 with the combination of T1 and T2-weighted MRI images, respectively.We constructed a model to predict the PALN metastasis in patients with cervical cancer using pre-treatment MRI image-based radiomics and machine learning. The model based on T2-weighted image or combination of T1 and T2-wighted MRI image showed promising prediction accuracy. This model may be useful to select patients who benefit from prophylactic EFI.I. Nishibuchi: None. D. Kawahara: None. M. Kawamura: None. K. Kubo: None. N. Imano: None. Y. Takeuchi: None. A. Saito: None. Y. Murakami: None. Y. Nagata: None.