Diffusion Kurtosis Imaging-Based Radiomics for Preoperative Prediction of Lymph Node Metastasis in Colon and Rectal Cancer.

RATIONALE AND

OBJECTIVES

To develop a diffusion kurtosis imaging (DKI)-based clinical-radiomics model for the preoperative prediction of lymph node metastasis (LNM) in colon and rectal cancer.

MATERIALS AND

METHODS

204 treatment-naive colorectal cancer (CRC) patients (108 colon, 96 rectal) were included. Radiomic features were extracted from apparent diffusion coefficient (ADC), mean diffusivity (DKI_D), and mean kurtosis (DKI_K) maps of primary tumors. Least absolute shrinkage and selection operator regression was used for feature selection. Logistic regression (LR), radial basis function support vector machine, and Naive Bayes were used to construct the ADC, DKI_D, DKI_K, and DKI models. LR was performed to construct the clinical and Combined models. The model performance was evaluated using receiver operating characteristic (ROC) analysis, calibration, and decision curve analysis. The model interpretability was assessed using Shapley additive explanations (SHAP).

RESULTS

The Combined model, integrating the DKI-derived radiomics score with maximum short-axis diameter of regional lymph nodes, achieved the highest discrimination, with area under the ROC curve (AUC) of 0.939 (95% CI: 0.906-0.968) in training and 0.957 (95% CI: 0.906-0.996) in testing, outperforming single-parameter models (p < 0.05). Subgroup analyses showed high performance in colon (AUC 0.980) and rectal (AUC 0.932) cancer. SHAP analysis confirmed the added value of DKI features.

CONCLUSION

The Combined model provides promising noninvasive prediction of LNM in colon and rectal cancer, aiding individualized risk stratification and treatment planning.