A Radiomic Approach To Predict The EGFR-TKI Therapy Response Of Non-small Cell Lung Cancer Patients

Part one Images Features in Rapid Progression Patients after EGFR-TKI therapy Purpose:To explore the specific pre-therapy CT conventional features for the first-line EGFR-TKI therapy NSCLC patients with rapid progression,in order to realize personalized therapy.Materials and Methods:From September 2009 to December 2018,a total of 188 patients with EGFR-mutant advanced non-small cell lung cancer who underwent first-line EGFR-TKI therapy were conducted.Response evaluation criteria in solid tumors(RECIST)1.1 was used to evaluate the TKI therapy response.The follow-up imaging and laboratory text were used to define the PFS.The follow-up interval should be 4-6 weeks.The clinical features including:age,gender,EGFR mutation and treatment.A total of 13 conventional CT features were evaluated by two radiologists including anatomic location,axial location,the largest axial diameter,boundary,air bronchogram,lymphangitic carcinomatosis,obstructive pneumonia,obstructive atelectasis,mediastinal enlarged lymph nodes,hilar enlarged lymph nodes,pleural effusion,pleural thickening,opposite lung suspected metastasis.Spearman correlation was used to explore the relation between the largest axial diameter and the PFS.Results:The PFS in all patients are 0.7-56.7(14.9±10.7)months.Short PFS was associated with central tumors(P=0.025),the largest axial diameter(r=-0.227,p<0.05),lymphangitic carcinomatosis(P=0.006),mediastinal enlarged lymph nodes(P<0.001)and hilar enlarged lymph nodes(P<0.001).The results of multiple linear regression showed that R is 0.52,R square is 0.27.The model is statistically significant(F=3.60,P<0.001).The independent variables with statistical significance includes age(B=0.178,P=0.027),the largest axial diameter(B=-1.139,P=0.008),mediastinal enlarged lymph nodes(B=-3.456,P=0.048),hilar enlarged lymph nodes(B=-3.693,P=0.039),opposite lung suspected metastasis(B=-4.306,P=0.010).Conclusions:The pre-therapy CT conventional features are associated with the prognosis in advanced NSCLC patients who had EGFR mutation and underwent first-line EGFR-TKI therapy.And the central tumors,mediastinal and hilar enlarged lymph nodes and opposite lung suspected metastasis are associated with short PFS.Part two A Radiomic Approach to Predict the PFS of non-Small Cell Lung Cancer with EGFR-TKI TherapyPurpose:To predict PFS of the EGFR-TKI therapy for improving the pre-therapy personalized management.Materials and Methods:From September 2009 to December 2018,a total of 188 patients with EGFR-mutant advanced NSCLC who underwent first-line EGFR-TKI therapy were conducted.The clinical TNM was ?B to ?.Response evaluation criteria in solid tumors(RECIST)was used to evaluate the TKI therapy response.A total of 1232 radiomics features were extracted by the PyRadiomics tool from the half-automatic delineated three-dimensional volume of interest(VOI)on the pre-therapy CT images.And 17 conventional CT features were collected including the 14 from the first part and another three as:the number of hilar and mediastinum enlarged lymph nodes and the deepest pleural effusion.Clinical factors were collected as well.Then the normalized features were imported to the feature selection procedure,including F-test,Cluster and Recursive Feature Elimination(RFE).Light Gradient Boosting Machine(LightGBM)regression model was built based on the selected features to predict the PFS.5 times 5-fold cross-validation methods were used for verification.Mean absolute error(MAE),mean squared error(MSE)and R square were used to evaluate the model prediction effect.Results:A total of 250 radiomic features were selected by the F-test.The Cluster selects 64 features.After the RFE,9 features were selected and used in the LightGBM regression model.In the validation set,the radiomic model achieved a result as MAE:35.26,MSE:4.83;R square score:0.656,respectively.For the combined model,the number of features after three steps of selection was changed from 250 to 113 and to 110.It showed good prediction performance for the PFS in both training and validation set(MAE:9.32 and 10.27;MSE:1.88and 2.31;r square score:0.875 and0.733,respectively).Conclusions:Pre-therapy CT radiomic features showed good prediction effect on the EGFR-TKI therapy response for the advanced EGFR-mutant non-small cell lung cancer patients.And combining with clinical features,the combined model can achieve the personalized prediction.It helps improve the pre-therapy personalized management.