A nomogram based on dual-layer spectral detector CT-derived 40-KeV virtual monoenergetic images and iodine maps for preoperative prediction of regional lymph node metastasis in colorectal cancer

To develop and validate a dual-layer spectral detector CT (DLSCT)-based nomogram integrating DLSCT-derived quantitative parameters, radiomics features extracted from 40-KeV virtual monoenergetic images (VMI) and iodine maps, and clinical factors for preoperative prediction of regional lymph node metastasis (LNM) in colorectal cancer (CRC). A total of 225 patients with pathologically confirmed CRC were retrospectively enrolled and randomly assigned to a training cohort (n = 157) and a validation cohort (n = 68) at a 7:3 ratio. Clinical and pathological data, as well as DLSCT-derived quantitative parameters, were collected. Radiomics features were extracted from DLSCT-derived 40-KeV VMI and iodine maps in both the arterial phase (AP) and venous phase (VP). Radiomics features were further selected using least absolute shrinkage and selection operator (LASSO) regression. Multiple machine learning algorithms were then applied to develop radiomics models and compare predictive performance. Logistic regression analysis was performed to identify independent clinical variables and DLSCT-derived quantitative parameters and to establish a clinical model. The radiomics score (Radscore), defined as the predicted probability from the optimal radiomics model, was integrated with clinical variables and DLSCT-derived quantitative parameters to construct a nomogram. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Serum carbohydrate antigen 19 − 9 (CA19-9), enhancement pattern, normalized iodine concentration in the AP (NICAP), CT value at 100-KeV in the VP (CT100KeVVP), and normalized iodine concentration in the VP (NICVP) were identified as independent predictors of LNM in CRC. The nomogram integrating the Radscore with clinical variables and DLSCT-derived quantitative parameters achieved the best predictive performance, with areas under the ROC curve (AUCs) of 0.959 and 0.920 in the training and validation cohorts, respectively. DCA indicated that the nomogram provided a higher net clinical benefit for predicting LNM in CRC. A DLSCT-based nomogram incorporating radiomics features derived from 40-KeV VMI and iodine maps may provide a useful tool for the preoperative prediction of regional LNM in CRC and could potentially aid individualized treatment planning.