Radiomics based on RESOLVE ADC in differentiating benign and malignant lymph nodes in rectal cancer

Background: The preoperative prediction of lymph node (LN) metastasis (LNM) in rectal cancer assumes a pivotal role in risk stratification and therapeutic decision-making. This study aimed to construct and validate a clinical-radiomics nomogram in non-invasively predicting LNM based on readout segmentation of long variable echo-trains (RESOLVE) apparent diffusion coefficient (ADC). Methods This retrospective study included preoperative images of 206 rectal cancer patients, containing 261 pathologically confirmed LNs, at our hospital between August 2018 and February 2022. There were 144 patients (55 in the LNM group and 89 in the non-LNM group) with 182 LNs (76 metastatic LNs and 106 benign LNs) in the training cohort and 62 patients (25 in the LNM group and 37 in the non-LNM group) with 79 LNs (31 metastatic LNs and 48 benign LNs) in the validation cohort. Image segmentation was performed by manually delineating the whole tumor and the maximum cross-section of each LN. The maximum-relevance and minimum-redundancy (mRMR) and least absolute shrinkage and selection operator (LASSO) methods were used to construct radiomics signature. Logistic regression modeling was employed to construct models based on clinical factors and LN morphologic criteria (model 1), tumor radiomics features combined with LN radiomics features (model 2), and model 2 combined with model 1 (model 3). Then model 3 was converted as a form of clinical-radiomics nomogram. Diagnostic performance was assessed by the area under the curve (AUC). The DeLong test was conducted to compare AUCs between models. Decision curve analysis (DCA) was performed to assess the clinical usefulness of three models. Results Both model 2 and model 3 showed higher AUCs in the training (model 2 0.930, model 3 0.948) and validation cohorts (model 2 0.864, model 3 0.887) than model 1(training 0.742; validation 0.745). Model 3 showed improved diagnostic performance over model 2 (P = 0.049) and model 1(P < 0.001) in the training cohort. The DCA indicated that model 3 was regarded as the candidate model to identify LN status. Conclusions The clinical-radiomics nomogram, incorporating RESOLVE ADC based-radiomic features from tumors and LNs with clinical factors and LN morphologic criteria was promising for predicting LNM in rectal cancer.