Based On Radiomics Of Non-small Cell Lung Cancer EGFR Mutation Prediction Model

In the first-line treatment of non-small cell lung cancer(NSCLC)patients,detection of epidermal growth factor receptor(epidermal growth factor receptor,EGFR)gene mutation status has become the pre-conditions decision whether patients can apply EGFR tyramine(EGFR-tyrosine kinase inhibitor,EGFR-TKI)drug treatment.However,there are still some challenges to the detection of EGFR mutation in patients' willingness,the way of obtaining tissue samples(biopsy,surgical resection,etc.),quality control of tissue samples,cost of collection,and heterogeneity of lung cancer.As a non-invasive,high-resolution rapid imaging modality,CT also plays an important role in the detection and diagnosis of lung cancer.The purpose of this study is to investigate whether there is correlation between CT image features of lung cancer and EGFR gene mutations,and to establish a prediction model of EGFR gene mutation based on CT image features.The main contents of this paper are as follows:(1)In this paper,110 cases of lung cancer CT image data from the Sheng jing Hospital were collected and analyzed according to the guidance of the radiologist to confirm the tumor area and related important features.The region of lung cancer was segmented by using adaptive region growth method,and the extraction of lung cancer region was completed.(2)Secondly,this paper used Haralick characteristic calculation method and LBP feature calculation method respectively to extract a total of 44 texture features to avoid the poor influence of external features of tumor based on the research results of the past scholars in the field of lung cancer feature extraction.Twenty-six texture features were extracted by Haralick calculation method,and eighteen texture features were extracted by LBP feature calculation method,and these features were prepared for further classification.(3)This paper had collected 110 cases of CT image data,and divided into 5 groups,choose one group as the test set,the other four groups as the training set;Then this paper used SVM algorithm?BP neural network algorithm and LEM algorithm to train and test the 44 texture feature,and established three prediction model for non-small cell lung EGFR mutations;Finally,this paper evaluated and analyze these three methods of classification.From the analysis result,it can be concluded that that ELM algorithm can get the best classification effect for predicting the relationship between EGFR mutations and CT image texture feature,and the prediction accuracy of the BP neural network algorithm is better than SVM algorithm but worse than ELM algorithm,the SVM classification get the worst effect among the three classifications.Therefore,a prediction(or classification)model of EGFR mutations based on CT image features(mainly texture features)of patients with lung cancer,may identify EGFR mutations in lung cancer patients and can help clinicians decide whether to use EGFR-TKI drug therapy.