Application Of CT Radiomics Features To Predict The EGFR Mutation Status And Therapeutic Sensitivity To TKIs Of Advanced Lung Adenocarcinoma

BackgroundLung cancer is considered the leading cause of death among cancer patients worldwide,and lung adenocarcinoma is the most common histological type of lung cancer.About 80%-85%of all lung malignancies are non-small cell lung cancers(NSCLC),and more than 70%of NSCLC are already locally advanced or have distant metastases when diagnosed.With the development of tumor molecular biology and genomics,molecularly targeted drugs have been widely used in the field of lung cancer.Epidermal growth factor receptor is the most common driver gene in non-small cell lung cancer and is one of the major targets that can effectively treat NSCLC.Epidermal growth factor receptor tyrosine kinase inhibitors have significantly improved prognosis in lung adenocarcinoma patients with(Epidermal growth factor receptor,EGFR)-sensitive mutations.Several studies have shown that EGFR-sensitive mutations are an important predictor of the effectiveness of tyrosine kinase inhibitors.EGFR tyrosine kinase inhibitors have been used as the standard first-line treatment for advanced EGFR-sensitive mutant non-small cell lung cancer.The traditional method to determine EGFR mutation status is pathological examination after tissue puncture biopsy,which can only provide tumor information at a single time point and a single spatial location due to the general condition of the patient,the small size of the tissue specimen,and the invasive nature of the biopsy.Patients with EGFR-sensitive mutations in non-small cell lung cancer can be treated with EGFR-TKIs with an efficiency of up to 70%,but 20%-30%of patients with EGFR-sensitive mutations are still primary resistant to GFRTKIs,and the practical application of EGFR-TKIs is to some extent limited by their primary resistance in clinical practice.limitations.Most of the predictive markers proposed in clinical practice are based on pre-and post-treatment expression changes to assess efficacy,and there is a lack of effective and clinically valid predictive markers that can accurately identify efficacy response before treatment to predict the efficacy of EGFR-TKIs therapy.Therefore,if patientswith primary resistance to EGFR-TKIs can be screened,it is of great scientific significance and clinical value to improve their efficacy and mitigate drug toxic side effects for the precision treatment of targeted lung cancer drugs.In the current context of precision medicine,finding a method that can predict gene mutation status,as well as predict early treatment efficacy and avoid treatment risks to effectively guide individualized treatment and maximize the prognosis of patients is an urgent problem in current clinical work.Artificial intelligence has developed rapidly in recent years,making its application in intelligent analysis of medical images increasingly widespread.Imaging histology is used to acquire a large amount of image data(CT,MRI,PET/CT),etc.,and extract a large number of quantitative imaging features,analyze and screen relevant imaging histological features,and build clinical prediction models for precise tumor treatment.This study applied to evaluate two major challenges in clinical targeting of non-small cell lung cancer:gene mutation status and response to targeted therapy,and construct a predictive model for targeted therapy in non-small cell lung cancer patients to achieve individualized and precise treatment of tumors and provide an auxiliary tool for clinical treatment decisions.Part Ⅰ:Prediction of advanced lung adenocarcinoma on EGFR expression status based on CT imaging features Purpose of the study.PurposesTo explore the correlation between enhanced CT-based imaging features and genotype,and to establish a model for predicting EGFR mutation status;to use imaging features and clinicopathological features to screen for features that can predict EGFR mutation status,and to construct a comprehensive prediction model for EGFR mutation in non-small cell lung cancer.Methods.One hundred and seventy patients with pathologically confirmed stage ⅢB or Ⅳ lung adenocarcinoma who underwent EGFR mutation testing were randomly selected from those enrolled in the retrospective analysis,and the study population was divided into two groups:(1)130 cases in the training group used to construct the training model(EGFR mutation=64,EGFR negative=66);(2)40 cases in the validation group used to construct the verification model(EGFR mutation=20,EGFR negative=20);baseline clinicopathological information was obtained from the cases,including:age,gender,smoking,TTF-1,NaspinA expression,etc.The CT imaging data of patients in the plain,arterial,and venous phases before treatment were selected,and the tumor areas were modified layer by layer by a radiologist with 5 years of experience in the GTVreference mediastinum and lung window automatically outlined by the MIM software and examined by a clinician with 10 years of experience in the GTV.Areas of disagreement were agreed upon after deliberation.According to the data needs of this paper,3D Slicer image processing software was applied to extract the image features of the lung adenocarcinoma tumor region.The Lasso-logistic regression model and 10-fold crossvalidation method were applied to analyze the three-stage temporal data to screen the imaging features associated with EGFR mutations in patients with non-small cell lung cancer.Finally,the selected features were used to construct feature labels for predicting EGFR mutations in patients with advanced lung adenocarcinoma.The efficacy of predicting EGFR mutation status in advanced lung adenocarcinoma was assessed clinically realistically throughsensitivity,specificity and accuracy by analyzing and validating multi-factor logistic regression models and constructing prediction models that fused the imaging histological feature labels and clinicopathological factors.Results.1.a model based on the imaging histological features of the flat-scan Phase predicted EGFR mutation status with an AUC value of 0.7625,sensitivity and specificity of 75.0%and 65.0%,respectively.2.a model based on the imaging histological features of the arterial Phase predicted EGFR mutation status with an AUC value of 0.8065,sensitivity and specificity of 80.0%and 80.0%,respectively.3.The model developed based on the imaging histological features of the venousphase predicted EGFR mutation status with an AUC value of 0.8075,sensitivity and specificity of 80.0%and 80.0%,respectively.4.The image features with predictive potential in the EGFR mutant group were 5 in the planar phase(1 neighboring grayscale difference matrix feature,4 wavelet transform features,AUC values 0.614-0.663,mean value 0.636),18 in the arterial phase(2 first-order features,1 grayscale level band matrix feature,15 wavelet transform features,AUC values 0.609-0.696,mean value 0.647),and 23 in the venous phase(1 first-order feature,1shape feature,1 neighborhood gray-level difference matrix feature,20 wavelet transform features,AUC values 0.603-0.700,mean 0.654).5.The comprehensive prediction model established by imaging histological features combined with clinicopathological features predicted EGFR mutation status with an AUC value of 0.920,sensitivity and specificity of 80.0%and 95.0%,respectively.ConclusionIn this study,a comprehensive prediction model combining imaging information(morphological basis features,first-order statistical features,grayscale co-occurrence matrix features,grayscale running length matrix features,and grayscale region size matrix features)and clinical features can better predict EGFR mutations and provide decision support for the treatment options of advanced non-small cell lung cancer by performing imaging histology analysis on chest CT of patients with initial diagnosis of advanced lung adenocarcinoma.The model can provide decision support for the selection of advanced NSCLC treatment options.Part Ⅱ:Predicting the sensitivity of advanced lung adenocarcinoma to EGFR-TKIs treatment based on CT imaging featuresPurposesTo help clinicians develop targeted treatment regimens for patients with advanced lung adenocarcinoma,imaging histology feature labels based on pretreatment CT images were constructed and their value in predicting the treatment sensitivity of TKIs was evaluated.To construct a model of patients’ TKIs treatment sensitivity by integrating imaging histological features and inflammatory factors indexes to achieve individualized treatment assessment of tumors and provide an aid for clinical treatment decision.MethodsFrom the patients enrolled in the retrospective analysis,167patients with pathologically confirmed stage ⅢB or Ⅳ lung adenocarcinoma,tested for EGFR mutations,and CT examinations 2 weeks before and after(within 3 months)treatment with oral TKIs,combined with patient imaging and laboratory tests,were randomly selected by an attending physician with 5 years of radiology diagnosis and an oncology associate with more than 10 years of experience using the RECIST 1.11.1 criteria to confirm the patient’s primary drug-resistant patients.Clinical information included:age,sex,height,weight,absolute neutrophil value,absolute lymphocyte value,serum albumin,nflammatory factors including:platelet to lymphocyte ratio(PLR),advanced lung cancer inflammatory index(ALI),ALI=BMI(kg/m2)× Alb(g/dL)/NLR,prognostic nutritional index(PNI),systemic immune inflammatory index,and neutrophil to lymphocyte ratio(NLR).The MIM Maestro 6.7.6(MIM Software,Inc.,USA)software was used to correct CT images of multiple phases(pan phase,venous phase)using a non-rigid alignment method with arterial phase images as a reference,and the MIM software automatically outlined the GTV of the whole tumor lesion on the lung window according to the threshold,and the study subjects were divided into two groups:(1)127 cases in the training group used to construct the training model(EGFR sensitive=The subjects were divided into two groups:(1)127 cases in the training group(EGFR sensitive=63,EGFR primary resistance=64)for constructing the training model;(2)40 cases in the validation group(EGFR sensitive=20,EGFR primary resistance=20)for constructing the validation model;the tumor target area imaging features were extracted by 3D Slicer software.The features extracted from the imaging histology were analyzed by applying Lasso-logistic regression model and 10-fold cross-validation method to build a model for screening patients with primary resistance to EGFR-TKIs.A comprehensive prediction model incorporating imaging histological features and inflammatory factors was constructed by multifactorial Cox regression model analysis.Results.1.A model based on the imaging histological characteristics of the flat-scan predicted TKIs with a sensitivity AUC of 0.73,sensitivity and specificity of 70.0%and 65.0%,respectively.2.A model based on the imaging histological characteristics of the arterial phase predicted TKIs with a sensitivity AUC of 0.8325,sensitivity and specificity of 65.0%and 70.0%,respectively.3.The model developed based on the imaging histological characteristics of the venous phase predicted TKIs with a sensitivity AUC value of 0.895,sensitivity and specificity of 65.0%and 90.0%,respectively.4.The image features with predictive potential in the TKIs treatment-sensitive group were 3 in the plain phase(3 wavelet transform features,AUC values 0.619-0.688,mean 0.649),7 in the arterial phase(1 neighborhood grayscale difference matrix,6 wavelet transform features,AUC values 0.606-0.686,mean 0.640)and 22 in the venous phase(2 first-order features,neighborhood grayscale 1 difference matrix feature,19 wavelet transform features,AUC values 0.616-0.703,mean 0.647.5.The integrated prediction model established by imaging histological features combined with inflammatory factor features had an AUC value of 0.999 for the treatment sensitivity of TKIs.ConclusionIn this study,we performed imaging histological analysis of chest CT before targeted therapy in patients with initially diagnosed advanced lung adenocarcinoma,and constructed an imaging histological model to predict patients’ treatment sensitivity to TKIs.The imaging histological features were effective in distinguishing patients with primary drug resistance.A comprehensive clinical prediction model based on the imaging histology label and inflammatory factors was constructed,which could more accurately predict patients with advanced lung adenocarcinoma sensitive to TKIs and provide a basis for guiding individualized patient treatment.It provides safe,non-invasive,effective and economical evidence for precision targeted therapy for advanced lung cancer.