A Radiomics Model Study Of Lymph-node Regression After Neoadjuvant Chemoradiotherapy For Rectal Cancer

Purpose: Lymph node(LN)response after neoadjuvant chemoradiotherapy(n CRT)is critical for individualized medical decision-making in locally advanced rectal cancer(LARC).To provide a reliable basis for decision-making,this study investigated the feasibility and efficacy of LN-based radiomics analysis to predict LN response after n CRT in LARC patients.Materials and Methods: In this study,78 patients with clinical stage T3-T4,N1-2,and M0 rectal adenocarcinoma were enrolled,all underwent neoadjuvant chemoradiotherapy(n CRT)and total rectal mesenteric resection with lateral suspected lymph node dissection.Pathologists assigned lymph-node regression grade(LRG)to the resected LNs and divided them into two groups based on the scoring results: pathological complete response(p CR)and non-p CR groups.A total of 243 LNs were included after strict inclusion and exclusion criteria,and they were allocated to a training cohort(n=173)and a validation cohort(n=70)in a 7:3 ratio.Two radiologists outlined layer-by-layer regions of interest(ROI)in the LN on T2 WI images before n CRT and extracted 3641 radiomics features from each LN.Firstly,the Mann-Whitney U test was used to initially screen features for correlation with LRG(P<0.05).Secondly,features were further screened using the least absolute shrinkage and selection operator(LASSO),and the radiomics signature model was constructed by calculating the radiomics score(Rad-score)of each LN.Meanwhile,LN morphology features related to LRG were screened out by univariate and multivariate analysis,and the LN morphology model was constructed using logistic regression.Finally,the screened LN morphological features were integrated with the Rad-score to construct a radiomics nomogram.The predictive performance of the three models was evaluated by the area under the curve(AUC)of the receiver operating characteristic curve.The calibration curve and decision curve analysis(DCA)calibrated and validated the models’ predictive performance.Results: Through comparison of the three models,it was found that the LN morphology model performed the worst(with AUCs of 0.757 and 0.823 for the training and validation cohorts,respectively),followed by the radiomics signature model(with AUCs of 0.908 and 0.865 for the training and validation cohorts,respectively),while the radiomics nomogram exhibited the highest predictive performance(with AUCs of 0.925 and 0.918 for the training and validation cohorts,respectively).The calibration curves show good discrimination and calibration of the radiomics nomogram in both the training and validation cohorts.The DCA confirms the value of the radiomics nomogram for clinical application.Conclusion: The nodal-based radiomics model can effectively predict the response of LARC patients to treatment with LNs after n CRT,which will help to develop individualized treatment plans and guide the implementation of non-operative strategies for LARC patients.