| Objective: Based on the chest CT images of patients with non-small cell lung cancer before chemoradiotherapy,the radiomics prediction model was constructed and verified to predict the partial response probability of patients receiving sequential chemoradiotherapy(SCRT)or concurrent chemoradiotherapy(CCRT)for non-small cell lung cancer,so as to provide reference for clinical treatment decision.Methods:The clinical treatment information and imaging data of patients diagnosed with non-small cell lung cancer and receiving SCRT or CCRT in the Affiliated Tumor Hospital of Anhui University of Technology from January 2016 to June 2020 were retrospectively collected.According to the sodium exclusion criterion,data of 75 patients with non-small cell lung cancer were collected,and according to the random seed number,according to the ratio of 7: 3,52 patients in the population were included in the training set and 23 patients were included in the internal verification set.A total of 30 patients who met the inclusion criteria from July 2020 to June 2021 were included in the time validation set.In the training set data,the imageological features with high stability were selected by Intraclass Correlation Co eficient(ICC).Then,the feature filtering for model construction was performed through hypothesis testing and the Least Absolute Shrinkage and Selection Operator(LASSO)algorithm.Finally,logistic regression(LR),decision tree(DT)and Ada Boost classifiers were used to construct the model for predicting the efficacy of chemoradiotherapy based on the screened radiographic features.The area under receiver operating characteristic curve curve(AUC),sensitivity and specificity were used to evaluate the prediction accuracy and stability of the model.In addition,that visual selection of the prediction model is display using nomogram;The effectiveness of that predictive model in clinical application was teste using decision curve analysis.Result: A total of 105 patients were included,including 52 cases in the training group,23 cases in the internal verification group and 30 cases in the interval verification group.In the training group data,through the hypothesis test and LASSO regression analysis,six imageological features are screened out,and the prediction model is constructed using machine learning method.In the training set,the AUCs of LR,DT,and Ada Boost models were 0.919,0.773,and 0.832.The sensitivities were0.92,0.94 and 0.86;The specificity was 0.67,0.60,and 0.80.The AUCs for LR,DT,and Ada Boost models in the internal validation set were 0.795,0.723,and 0.638;The sensitivities were 0.81,0.87 and 0.56;The specificity was 0.71,0.57 and 0.71.In the time-period validation sets: AUC of LR,DT,and Ada Boost models were 0.832,0.856,and 0.830;The sensitivities were 1.00,0,89 and 0.84;The specificity was 0.63,0.82 and 0.82.The use of LR models to construct decision curves suggests that a risk threshold of 0.1–0.92 increases the net benefit for the patient.Conclusion:In this study,we constructed and validated a model that can predict the remission probability of non-small cell lung cancer patients after receiving SRCT\CCRT based on the imaging omics technology and machine learning classifier,and the performance of the prediction model based on logistic regression is superior to the decision tree and Adaboost model.Figure [7] Table [4] Reference [28]... |
PDF Full Text Request