Abstract
BACKGROUND: Extramural venous invasion (EMVI), tumor deposits (TDs), and peripheral nerve invasion (PNI) are high-risk pathological features in patients with rectal cancer (RC), but their preoperative assessment with conventional computed tomography (CT) is limited. This study aimed to evaluate the value of dual-energy computed tomography (DECT)-based fractal analysis for predicting EMVI/TD and PNI in patients with RC. METHODS: A total of 130 patients with histologically confirmed RC who underwent preoperative DECT were consecutively enrolled and divided into a retrospective development cohort (n=85) and a prospective validation cohort (n=45). Fractal dimensions (FDs) from conventional mixed-energy images (FD-Con), iodine maps [FD-iodine concentration (FD-IC)], and effective atomic number (Zeff) maps (FD-Zeff) were calculated. Patients were classified as PNI-positive/-negative and EMVI/TD-positive/-negative according to surgical pathology. Group comparisons, receiver operating characteristic (ROC) curve analysis, and multivariable logistic regression were used to identify independent predictors and build combined models, which were tested in the validation cohort. RESULTS: Among the 130 patients, 59 were PNI-positive and 35 were EMVI/TD-positive. FD-IC and FD-Zeff were significantly higher in positive than in negative groups for both PNI and EMVI/TD (all P values ≤0.019). In the validation cohort, FD-IC and FD-Zeff achieved areas under the ROC curves (AUCs) of 0.752 and 0.771 for predicting PNI and 0.795 and 0.855 for predicting EMVI/TD, respectively. FD-IC was an independent risk factor for PNI [odds ratio (OR) =77.873; P=0.039], and FD-Zeff was an independent predictor of EMVI/TD (OR =1.109×10(4); P=0.006). In the validation cohort, the combined model yielded AUCs of 0.771 and 0.879 for PNI and EMVI/TD, respectively. CONCLUSIONS: DECT-based fractal analysis, particularly FD-IC and FD-Zeff, provides quantitative markers for preoperative prediction of PNI and EMVI/TD in patients with RC and improves diagnostic performance as compared with conventional CT-based fractal parameters.