CT-based radiomics nomogram for the pre-operative prediction of lymphovascular invasion in colorectal cancer: a multicenter study

Abstract

Objective: To develop and externally validate a CT-based radiomics nomogram for the pre-operative prediction of lymphovascular invasion (LVI) in patients with colorectal cancer (CRC). Methods: 357 patients derived from 2 centers with pathologically confirmed CRC were included in this retrospective study. Two-dimensional (2D) and three-dimensional (3D) radiomics features were extracted from portal venous phase CT images. The least absolute shrinkage and selection operator algorithm and logistic regression were used for constructing 2D and 3D radiomics models. The radiomics nomogram was developed by integrating the radiomics score (rad-score) and the clinical risk factor. Results: The rad-score was significantly higher in the LVI+ group than in the LVI- group (p < 0.05). The area under the curve (AUC), accuracy, sensitivity and specificity of the 3D radiomics model were higher than those of the 2D radiomics model. The AUCs of 3D and 2D radiomics models in the training set were 0.82 (95% CI: 0.75–0.89) and 0.74 (95% CI: 0.66–0.82); in the internal validation set were 0.75 (95% CI: 0.65–0.85) and 0.67 (95% CI: 0.56–0.78); in the external validation set were 0.75 (95% CI: 0.64–0.86) and 0.57 (95% CI: 0.45–0.69); respectively. The AUCs of the nomogram integrating the optimal 3D rad-score and clinical risk factors (CT-reported T stage, CT-reported lymph node status) in the internal set and external validation set were 0.82 (95% CI: 0.73–0.91) and 0.80 (95% CI: 0.68–0.91), respectively. Conclusion: Both 2D and 3D radiomics models can predict LVI status of CRC. The nomogram combining the optimal 3D rad-score and clinical risk factors further improved predictive performance. Advances in knowledge: This is the first study to compare the difference in performance of CT-based 2D and 3D radiomics models for the pre-operative prediction of LVI in CRC. The prediction of the nomogram could be improved by combining the 3D radiomics model with the imaging model, suggesting its potential for clinical application.