Based On Radiomics And Clinical Predicted Model:A Method To Forecast Patients With Acute Paraquat Poisoning Who Will Have Respiration Failure

Objective: Establish and evaluate the efficiency of two clinical predicted models,which were based by mere radiomics and radiomics combined with clinical data to predict whether the acute paraquat poisoning patients will get respiratory failure.Methods: Collect and select the information of 75 patients from 2014 to 2017.The information include clinical data and chest computed tomography(CT),which were taken in the early stage.Randomly assign the patients to two group,whose names are training group and verification group using a 2:1 ratio.Use the information of training group(50 cases) to establish the prediction model,and use the information of verification group(25 cases) to evaluate the efficiency of model.Obtain the radiomics traits from the whole lungs CT of patients.Reduce the dimension of the radiomics traits by Principal component analysis(PCA) and lasso regression.Use only radiomics traits to establish a clinical predicted model(radscore) by the method of multivariate logistic regression analysis.Then use radiomics traits together with clinical stata to establish another clinical predicted model by the method of multivariate logistic regression analysis.Build the calibration curve of two models.Paint the nomogram to express the model whose has a better efficiency.Results: The radscore and the result of comprehensive model both show a statistically significant pertinence with the phenomenon that patient will get respiratory failure(P <0.001),in training group and verification group.In training group and verification group the areas under the ROC curve(AUC)of radscore are 0.8718(95%CI0.773-0.971) and 0.8333(95%CI0.655-1) respectively,at the same time the sensitivity and specificity respectively are 77%,80% and 89%,80%,and the accuracy of radscore are 84% and 80%.The predictors of comprehensive model include radscore WBC CK-MB and PQC.The AUC of comprehensive model is 0.9675(95%CI0.929-1) in training group.The sensitivity of the model is 86%,and the specificity of the model is 100%,meanwhile the accuracy of the model is 94% in training group.The AUC of comprehensive model is 0.9467(95%CI0.866-1)in validation group.The sensitivity of the model is 80%,and the specificity of the model is 100%,meanwhile the accuracy of the model is 84% in validation group.The calibration curve express that the comprehensive model shows a ideal efficiency.Conclusion: Radiomics shows a good application value to get information from chest computed tomography of the acute paraquat poisoning patients.The muscle damage has pertinence with lung damage in patients who have acute paraquat poison.Doctors and patients now have a new measure by the method of radiomics and establishing clinical predicted model to predict whether a patient will have respiratory failure or not,and the result will help doctors to recognized the critical patients and choose a proper therapy.