Abstract
This study aimed to evaluate the utility of a multiparametric MRI-based radiomics nomogram for identifying patients with rectal cancer (RC) at high risk of synchronous distant metastasis (SDM). A fusion feature selection strategy, which combined univariate analysis with three machine learning algorithms, was employed to optimize predictive signatures from the 1,688 radiomics features extracted using PyRadiomics. A retrospective cohort of 169 RC patients (stratified into training and test sets at an 8:2 ratio, n = 134/35) was analyzed. Among these, 48.5% (82/169) presented with SDM. Following the screening process, four clinical characteristics were selected. Feature selection yielded eight features from diffusion-weighted (DW) images, eight from T2-weighted (T2W) images, and six from the combined radiomics model (integrating DW and T2W phases). The clinical-radiomics nomogram demonstrated superior predictive performance over standalone clinical or radiomics models, achieving areas under the curve (AUC) of 0.93 (95% CI: 0.89-0.96) and 0.94 (95% CI: 0.79-0.97) in the training and test cohorts, respectively, with balanced sensitivity (0.85-0.88) and specificity (0.86-0.89). The calibration plots demonstrated alignment between the nomogram's predictions and actual outcomes. Decision curve analysis (DCA) indicated that the nomogram model provided the highest net benefit across both the training and test sets, outperforming the standalone clinical and radiomics models. Consequently, this MRI-based radiomics nomogram could assist in the preoperative identification of RC patients at high SDM risk, thereby optimizing clinical management strategies.