Prediction of Lymph Node Metastasis in Rectal Cancer Based on Super-Resolution MRI Radiomics and Clinical Baseline

Abstract

Abstract Objective To explore the clinical practical value of the super-resolution(SR) MRI radiomics model based on clinical baseline for predicting lymph node metastasis in rectal cancer before surgery. Methods Retrospective inclusion of 302 eligible patients with rectal cancer (109 with lymph node metastasis). Patients from one hospital were included in the training set (n = 181), while patients from other hospitals were included in the external validation set (n = 121). Super-resolution algorithm was developed to axial T2-weighted imaging (T2WI) and subsequent SR-T2WI images were generated. The conventional radiomics models and SR radiomics model were built by 8 machine learning algorithms separately, and the best model was selected as the radiomics model. Using single-factor and multivariate logistic regression analysis to identify clinical risk factors for building a clinical model, and combining it with the radiomics model to construct a joint model. Comparing the diagnostic efficacy of the three models using area under the curve (AUC) in ROC curves. Finally, comparing the diagnostic efficacy of the best predicted model with different experienced radiologists. Results After feature screening and dimension reduction, 5 and 10 radiomics features were retained for conventional images and SR images, respectively. The diagnostic performance of the SR model on the external validation set was better than that of the conventional image model. Three clinical risk factors related to lymph node metastasis were screened to develop a clinical model. By combining SR radiomics features with clinical risk factors, a joint model was constructed, and compared with the three models, the joint model demonstrated the best diagnostic performance with an AUC, sensitivity, specificity and accuracy of 0.756 (95% confidence interval(CI): 0.658–0.854), 69.2%, 75.6%, and 73.6% on the external validation set, which was superior to that of a radiology expert with 36 years of experience (AUC, sensitivity, specificity, and accuracy of 0.679 (95% CI: 0.588–0.830), 84.6%, 51.2%, and 62.0%) on the external validation set (P = 0.02), indicating high clinical utility value. Conclusion The SR MRI radiomics model based on clinical baseline has high clinical practical value in predicting lymph node metastasis before surgery of rectal cancer.