Establishment Of A Model Of Defining Benign And Malignant Ground Glass Opacities Based On The Feature Collected From Quantitative CT Images

Objective: More and more patients have found ground glass opacity(GGO)in the lungs by computed tomography(CT).GGO is a non-specific manifestation of the cause,including tumors,infection,local bleeding or interstitial fibrosis,have caused difficulties in clinical diagnosis and treatment.Imaging radiomics is a new subject that is rapidly developing.It can quantitatively analyze patients’ CT images through non-invasive means.By exploring the data information of radiography,it can help clinical diagnosis and treatment.The purpose of this study was to analyze the imaging features of GGO on the thoracic CT before surgery,and to establish a model for the diagnosis of benign and malignant GGO,which would provide some help for the diagnosis of patients with clinical manifestations of GGO.Methods: A retrospective analysis of 139 patients with benign and malignant lung tumors who underwent surgical resection from the First Affiliated Hospital of China Medical University from July 2014 to July 2018 was performed to extract CT images of the lungs before surgery.The CT lesions were delineated,and the features were extracted,dimensioned,analyzed and modeled by imaging radiomics.The radiography label was constructed to predict the lesions on lung CT as benign and malignant GGO.Results: Patients with benign and malignant tumors were randomly divided into a training set and a validation set.Based on the preoperative CT image of the lung,the image features of the patient were extracted.The image features were extracted including first-order statistical features such as intensity,shape,texture and wavelet.Second-order statistical features and high-order statistical features were extracted,and a total of 592 imaging radiomics quantitative features were extracted.After LASSO regression,the high-throughput image features were subjected to dimensionality reduction,and six imaging features were obtained to construct the predictive model.The signature was constructed in the training set and then verified in the test set.The sensitivity of the imaging ensemble model to the GGO benign and malignant judgment is 68.6% and the specificity is 65.7%.The GGO is benign and malignant in the verification set.The sensitivity of the judgment was 65.7% and the specificity was 67.6%.The clinicalcharacteristics of all patients included were analyzed.Gender(p=0.004)and smoking history(p=0.047)were significantly different between benign and malignant nodules.The remaining age(p=0.054),family history of tumor(p= 0.373)and family history of lung cancer(p=0.23)were not significantly different.The training model was under the ROC curve.The Area Under Curve(AUC)is 0.743,and the AUC of the verification concentration model is 0.702.There were 52 male patients(37.41%)and 87 female patients(62.59%).The patients in the training group were 21-79 years old,including 29 male patients(41.43%)and 41 female patients(58.57%).The validation patients were aged 37-78 years old,including 23 male patients(33.33%)and 46 female patients(66.67%).Conclusion: Based on the characteristics of lung CT images of patients before surgery,it is possible to construct a model of benign and malignant GGO,which can help patients with clinical diagnosis and treatment.