Radiomic Features Based On MRI And Pathology As Predictors Of Pathological Complete Response To Neoadjuvant Chemoradiotherapy In Locally Advanced Rectal Caner

Purpose To evaluate the value of delta-radiomic features(DRFs)for predicting pathological complete response(pCR)in patients with locally advanced rectal cancer(LARC).Materials and Methods In this retrospective study,a total of 172 patients with LARC were included and divided into a training set(n=121)and a test set(n=51).All patients received nCRT before surgery and were examined using MRI before and after nCRT.For radiomics,one radiologist manually delineated all regions of interest(ROIs)on pre-and post-nCRT T2-weighted imaging(T2WI),and 1049 RFs were extracted from pre-and post-nCRT MRI,respectively.DRF was defined the change in RF from pre-nCRT to post-nCRT RF,including the absolute change in RF(△-RF),and the relative change in radiomic feature(△-RF%).Feature selection methods including the least absolute shrinkage and selection operator algorithm(LASSO)were used to choose the optimized features for building the △-RF and △-RF%model.Logistic regression(LR)was used to construct the DRF combined model.For conventional MRI evaluation,one experienced radiologist assessed conventional radiological parameters such as MRI-based tumor regression grade(mrTRG),and a clinical plus conventional radiological model was built.A combined model including clinical,conventional,and DRF information was constructed.The diagnostic performance of different models was evaluated by receiver operator characteristic(ROC)curve analysis.Net reclassification index(NRI)was preformed to compare the predictive value of pCR between the clinical plus traditional radiological model and combined model.Results There were 27(15.7%)patients achieved pCR.In the △-RF and △-RF%model,7 most relevant RFs were retained,respectively.The △-RF and △-RF%model,and DRF combined model yielded AUCs of 0.838-0.853 and 0.799-0.831 in the training and test sets.The AUCs of clinical plus conventional radiological model were 0.804 and 0.802 in the training and test sets.The combined model achieved the highest AUC of 0.883(95%confidence interval[CI]:0.798-0.968)in the training set,and were successfully validated in the test set(AUC 0.875[95%CI:0.778-0.972]).Adding DRF information to clinical plus conventional radiological model significantly improved the predictive value for pCR(training set:NRI=0.692,p<0.001;test set:NRI=0.520,p=0.044).Conclusion Our study demonstrated that DRFs can improve pCR prediction when integrated with clinical and traditional radiological features and thus were promising to assist in clinical decision making for patients with LARC.Purpose To investigate the diagnostic performance of the patho-radiomic signatures from pathological and radiological features for predicting pathological complete response(pCR)after neoadjuvant chemoradiotherapy(nCRT)in patients with locally advanced rectal cancer(LARC).Materials and Methods A total of 153 patients with LARC were enrolled in this retrospectively study between January 2015 and June 2018.All patients underwent MR examination before and after nCRT,and all pathological sections were harvested from proctoscopic biopsy before nCRT.A total of 214 radiomic features were obtained from T2-weighted MR imaging(T2WI)and 512 pathomic features were extracted from each patient pathological section.Feature selection methods including the least absolute shrinkage and selection operator algorithm(LASSO)was used for building patho-radiomics signature,radiomics signature,and pathomics signature.Receiver operator characteristic(ROC)curve analysis was used to evaluate the diagnostic performance.Results Overall,17 patients(15.7%)and 7 patients(15.6%)achieved pCR in the training and test sets,respectively.The patho-radiomics signature,which comprised 8 selected features,was significant associated with pCR.The area under the ROCs(AUC)of the patho-radiomics signature for pCR prediction were 0.89(95%CI:0.83-0.96)and 0.86(95%CI:0.73-1.00)in the training and test set.The radiomics signature from joint pathological and radiological features outperformed the radiomics signature from either of them alone.Conclusion In LARC patients who underwent nCRT,the patho-radiomics signature created by combining pathological and radiological features exhibited favourable for pCR prediction and can increase the confidence of the organ-preserving strategy.