Related Experiment Video
Updated: Jun 12, 2026
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
Radiomics analysis of energy CT-derived iodine density maps for predicting lymph node metastasis in colorectal cancer
Ziqi Jia1, Danfeng Wang2, Anni Pan3
1Department of Radiology, The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, Guangzhou, China.
Background:
Accurate preoperative assessment of lymph node metastasis (LNM) is crucial for clinical staging and the development of personalized surgical strategies in colorectal cancer (CRC). This study aimed to investigate the efficacy of radiomics analysis utilizing iodine density maps (IDMs) from dual-layer detector spectral CT (DLCT) in the preoperative prediction of LNM in CRC patients.
Methods:
A retrospective analysis was conducted on 465 CRC patients from two centers, categorized into training (n=304), validation (n=80), and test sets (n=81). Following manual tumor segmentation, radiomics features were extracted separately from IDMs obtained during arterial and venous phases (AP, VP). Random forest models were developed for radiomics, and combined clinical-radiomics predictions, whereas the clinical model was constructed using multivariate logistic regression. The models' LNM predictive performance was assessed using receiver operating characteristic curves (AUC), calibration curves, and decision curves. SHapley Additive exPlanation (SHAP) analysis was utilized to interpret feature contributions.
Results:
A total of 1,888 radiomics features were extracted from the volume of interest (VOI) of iodine density (ID) images from dual phases. Totals of 20, 20, and 14 features were selected for constructing an AP radiomics model, a VP radiomics model, and a dual-phase radiomics model. The clinical-radiomics model, incorporating both the AP and VP features and clinical variables, demonstrated the highest predictive performance for LNM across all datasets, achieving AUC values of 0.948 [95% confidence interval (CI): 0.918-0.977], 0.921 (95% CI: 0.839-0.970), and 0.830 (95% CI: 0.731-0.905) in the training, validation, and testing sets. SHAP analysis indicated that wavelet-transformed texture and first-order features were significant contributors to LNM predictions.
Conclusions:
The clinical-radiomics model based on dual-phase IDMs is a noninvasive approach for preoperative prediction of LNM in CRC patients.
