Radiomic Analysis for Pretreatment Prediction of Recurrence After Radiotherapy in Locally Advanced Cervical Cancer

【Objective】 To predict the recurrence of advanced cervical cancer patients treated with radiotherapy from radiomics features on pre-treatment T1- and T2-weighted MRI images. 【Methods】 A total of 90 patients were split into two sets: 67 patients for model training and 23 patients for model testing. The patient outcome was classified into two groups; patients with a recurrence (group I) and without a recurrence (group II). The predictors were selected using the least absolute shrinkage and selection operator (LASSO) regression. The machine learning for the predictive models was sued neural network classifiers. The accuracy, sensitivity, specificity, and the area under the curve (AUC) from the receiver operating characteristic were evaluated. 【Results】 By the LASSO analysis, we found 25 radiomics features from the T1-weighted MRI image and 4 radiomics features from the T2-weighted MRI image. The accuracy of the prediction model was highest with the combination of T1- and T2-weighted MRI images. The model performances with T1-weighted MRI image and T2-weighted MRI image were 86.4% and 89.4% of accuracy, 74.9% and 38.1% of sensitivity, 81.8% and 72.2% of specificity, and 0.89 and 0.69 of AUC. The model performance was improved with the combination of T1- and T2-weighted MRI images, which was 93.1% of accuracy, 81.6% of sensitivity, 88.7% of specificity, and 0.94 of AUC. 【Conclusions】 The radiomics analysis with T1- and T2-weighted MRI images could highly predict the recurrence of the cervix cancer after radiotherapy. The variation of the distribution and the difference of the pixel number at the peripheral and the center were important predictors.