Development And Validation Of Models Based On Radiomics And Machine Learning For Prognostic Prediction In Patients With Malignant Biliary Obstruction Underwent Stent Placement

Part ? Risk Prediction for Early Biliary Infection after Percutaneous Transhepatic Biliary Stent Placement in Malignant Biliary ObstructionPurpose: To establish a nomogram for predicting the occurrence of early biliary infection(EBI)after percutaneous transhepatic biliary stent placement(PTBS)in malignant biliary obstruction(MBO)and evaluate the predictors by artificial neural network(ANN).Materials and Methods: Patients treated with PTBS for MBO were allocated to a training cohort or a validation cohort in this multicenter study.The independent risk factors for EBI selected by multivariate analyses in the training cohort were used to develop a predictive nomogram.An ANN model was applied to assess the importance of these factors.The predictive accuracy of this nomogram was determined by concordance index(c-index)and a calibration plot,both internally and externally.Results: A total of 243 patients(training cohort: n = 182;validation cohort: n = 61)were included in this study.The independent risk factors were length of obstruction(OR,1.061;95% CI,1.013-1.111;P = 0.012),diabetes(OR,5.070;95% CI,1.917-13.412;P = 0.001),location of obstruction(OR,2.283;95% CI,1.012-5.149;P = 0.047)and previous surgical or endoscopic intervention(OR,3.968;95% CI,1.709-9.217;P = 0.001),which were selected into the nomogram.According to the ANN model,length of obstruction indicated the largest importance among the four risk factors.The c-index values showed good predictive performance in the training and validation cohorts(0.792 and 0.802,respectively).The optimum cut-off value of risk was 0.25.Conclusion: The nomogram can facilitate the early and accurate prediction of EBI in patients with MBO who underwent PTBS.ANN model suggested the length of obstruction is an important predictor.Patients with high risk(> 0.25)should be administered more effective prophylactic antibiotics and undergo closer monitoring.Part II Early-warning models to estimate the 30-day mortality risk after stent placement for patients with malignant biliary obstruction: logistic model versus artificial neural network modelPurpose: To develop,validate and compare the early-warning models(a logistic regression model and an artificial neural network [ANN] model)of the 30-day mortality risk for patients with malignant biliary obstruction(MBO)undergoing percutaneous transhepatic biliary stent placement(PTBS).Materials and Methods: Between January 2013 and October 2018,this multicenter retrospective study included 299 patients with MBOs who underwent PTBS.The training set consisted of 166 patients from four cohorts,and another two independent cohorts were allocated as external validation sets A and B with 75 patients and 58 patients,respectively.The logistic and ANN models were developed to predict the risk of 30-day mortality after PTBS.The predictive performance of these two models was validated internally and externally.Results: The logistic model was developed based on two independent predictors,and the ANN model was developed based on eight candidate predictors with one hidden layer and one unit.The ANN model had higher values of area under the curve than the logistic model in the training set(0.819 versus 0.797)and especially in the validation sets A(0.802 versus 0.714)and B(0.732 versus 0.568).Both models had high accuracy in the three sets(75.9-83.1%).Along with a high specificity,the ANN model improved the sensitivity.The net reclassification improvement(12.0-16.5%)and integrated discrimination improvement(5.9-13.5%)also demonstrated that the ANN model led to improvements in predictive ability.Conclusions: Early-warning models were proposed to predict the risk of 30-day mortality after PTBS in patients with MBO.The logistic model is easier to use,while the ANN model has higher accuracy and better generalizability.Part III Radiomics facilitates restenosis-free survival prediction and risk stratification among patients with unresectable pancreatic cancer underwent irradiation stent placementPurpose: To develop a model combined with radiomics for restenosis-free survival(RFS)prediction and risk stratification among patients with unresectable pancreatic cancer with malignant biliary obstruction(UPC-MBO)undergoing irradiation stent placement.Materials and Methods: This retrospective study included 106 patients treated with an irradiation stent for UPC-MBO.These patients were randomly divided into a training group(74 patients)and a validation group(32 patients).A clinical model for predicting RFS was developed with clinical predictors selected by univariate and multivariate analyses.After integrating the radiomics signature,a combined model was constructed to predict RFS.The predictive performance was evaluated with the concordance index(c-index)in both the training and validation groups.The median risk score of progression in the training group was used to divide patients into high-and low-risk subgroups.Results: Radiomics features were integrated with clinical predictors to develop a combined model.The predictive performance was better in the combined model(c-index,0.791 and 0.779 in the training and validation groups,respectively)than in the clinical model(c-index,0.673 and 0.667 in the training and validation groups,respectively).According to the median risk score of 1.264,the RFS was significantly different between the high-and low-risk groups(P < 0.001 for the training group,and P = 0.016 for the validation group).Conclusion: The radiomics-based model had good performance for RFS prediction in patients with UPC-MBO who received an irradiation stent.Patients with slow progression should consider undergoing irradiation stent placement for a longer RFS.