Radiomics Combined With Clinical Characteristics Predicted The Progression-Free Survival Time In First-Line Targeted Therapy For Advanced Non-Small Cell Lung Cancer With EGFR Mutation

Background:Radiomics,obtaining information from images,is more and more studied in lung cancer.The aim of this study was to explore the most appropriate modeling method to construct a model of advanced non-small cell lung cancer of EGFR mutations that can predict the progression-free survival of EGFR-TKI treatment.Methods:A total of 100 patients with 19 del or 21L858 R mutations and received EGFR-TKIs as first-line treatment were included in our study.Radiomics features were extracted from pre-treatment CT images of patients.Then,our study used 22 feature selection methods and 8 classifiers constructed 176 discrimination models after arrangement and combination.The average performance of 176 models obtained from only radiomic features and radiomic features combine clinical features were compared and selected better performances method for subsequent model comparison.Each model was evaluated using AUC,ACC,sensitivity and specificity to obtain the best comprehensive performance model.Furthermore,the characteristics of the final inclusion model were compared between groups.The risk score of patients was calculated by the optimal model,and the patients were divided into high risk group and low risk group according to the risk score.Kaplan-Meier and log-rank test methods were used to evaluate and compare the survival curves of high risk and low risk groups.Results:107 radiomic features and 7 clinical features were obtained from each patient.After feature selection,the top-ten most relevant features were input into 176 models.The modeling process found that the performance of the identification model constructed by different feature selection methods and different machine learning methods was different,and the performance of the model was poor using only the top10 optimal influencing group features.The average AUC of all models constructed with radiomic features was 0.524 and that of all models constructed with clinical features plus radiomic features was 0.591(p=0.000).Therefore,this study used the clinical features combine radiomic features to construct the models.The top-ten features of models included three clinical factors(smoking,mutation and outcome),four shape-based features(elongation,least axis length,flatness and major axis length),one first-order based feature(interquartile range),and two texture based features(glcm-difference variance and glcm-small area emphasis).The Logistic regression(LR)model using gini-index feature selection acquired the best performance(AUC=0.797,ACC=0.722,sensitivity=0.758,specificity=0.693).Comparing top 10 features included in the model,with only two texture features,GLSZM(Small Area Emphasis)(p=0.003),GLCM(Difference Variance)(p=0.024),two clinical features,mutant genes(p=0.030)and outcome(p=0.000)were independent risk factors for PFS of EGFR-TKI therapy for advanced NSCLC with EGFR mutations.The median R-score was 0.518(IQR,0.023-0.987)as cut-off value divided the patients into high-risk and low risk groups.The KM survival curves of the two groups showed better stratification results(p=0.000).Conclusion:Our study suggests that Radiomics combined with clinical features can be used to predict progression-free survival of EGFR-TKI therapy of NSCLC.Selecting various modeling methods to filter out the optimal model can improve the accuracy of prediction.The best performance was obtained using gini-index LR models(AUC=0.797,ACC=0.722,sensitivity=0.758,specificity=0.693).The study found that GLSZM(Small Area Emphasis),GLCM(Difference Variance),mutation gene and efficacy evaluation were independent risk factors for progression-free survival in first-line targeted therapy for advanced NSCLC with EGFR mutation.