Preoperative Prediction of Lymph Node Metastasis in Rectal Cancer Using Multimodal Magnetic Resonance Imaging Radiomics Combined with Clinical Risk Factors.
Abstract
Rationale
AND
Objectives
This study aimed to investigate the preoperative predictive value of multimodal magnetic resonance imaging (MRI) radiomics combined with clinical risk factors for lymph node metastasis in rectal cancer, and to construct and validate an efficient and reproducible preoperative prediction model.
Materials And Methods
A retrospective analysis was conducted on 129 patients with pathologically confirmed rectal cancer who underwent surgery at our hospital between 2017 and 2023. Preoperative pelvic T2-weighted imaging (T2WI), contrast-enhanced T1-weighted imaging (T1CE), and complete clinical data were collected. Three-dimensional regions of interest were manually segmented on the tumors using ITK-SNAP software, and radiomic features were selected via variance filtering and least absolute shrinkage and selection operator (LASSO) regression. Patients were randomly divided into a training cohort (n = 90) and an internal validation cohort (n = 39) at a 7:3 ratio. Five Logistic regression prediction models were constructed: single-sequence radiomics, multimodal radiomics, multimodal radiomics combined with clinical risk factors, and clinical factors alone. Using postoperative pathology as the gold standard, the diagnostic performance of each model was evaluated by the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy.
Results
The multimodal radiomics combined with clinical risk factors model demonstrated the best predictive performance, with an AUC of 0.885 (95% CI: 0.819-0.937), sensitivity of 85.3%, specificity of 82.1%, and accuracy of 83.6% in the internal validation cohort. Its performance was significantly superior to that of the single-sequence models, the multimodal radiomics model, and the clinical model alone (all P < 0.05). Entropy from T2WI, peak enhancement from T1CE, tumor volume, tumor size, carcinoembryonic antigen (CEA), and neutrophil-to-lymphocyte ratio (NLR) were identified as independent predictors of lymph node metastasis. Calibration curves and decision curve analysis confirmed that the model was well-calibrated and provided substantial net clinical benefit.
Conclusion
Integrating multimodal MRI radiomics from T2WI and T1CE with clinical indicators such as CEA and NLR significantly improves the preoperative prediction of lymph node metastasis in rectal cancer, demonstrating promising clinical applicability. However, as this was a single-center retrospective study, further multicenter, large-sample external validation is warranted to confirm the stability and generalizability of the model.