Diagnostic Value Of CT-Radiomic Features For Histological Invasiveness In PGGNs Less Than Or Equal To 10 Mm

Part One Comparasions of CT findings and histological invasiveness inpGGNs less than or equal to 10 mmObjective: To investigate the correlation between CT findings and pathological invasion of pure ground-glass nodules(GGNs)less than or equal to 10 mm in size,and to improve the diagnostic accuracy of pure ground glass nodules.Materials and methods: This retrospective study included 102 patients with 102 pGGNs? 10 mm who were confirmed by surgery and pathology.There were altogether of 60 pre-invasive lesions(25 AAH and 35 AIS)and 42 invasive lesions(27 MIA and 15 IAC).CT finds of pGGNs including size,mean CT value,location,shape,tumor-lung interface,lobulation,spiculation,bubble luccency,air bronchogram,pleural indentation and vascular changes were evaluated.Results: There were significant differences in size,mean CT value and lobulation between preinvasive group and invasive group(P?<?0.05),while no significant differences in age,gender,shape,tumor-lung interface,bubble lucency,air bronchogram,pleural indentation,and vascular convergence or dilatation were found(P?>?0.05).ROC analyses revealed that the optimal cut-off value for discriminating pre-invasive from invasive lesions was 8.7 mm for size and for mean CT value was-572 HU.Conclusion: The lesion size,mean CT value and lobulation can help differentiate pre-invasive lesions and invasive lesion appearing as pure ground-glass nodules ?10 mm.A maximum diameter ?8.7 mm or a mean CT value greater than-527 HU indicate MIA or IAC.Part Two Predictive value of Radiomic features of p GGNs less than or equal to 10 mm for histological invasivenessObjective: Radiomics can quantify tumor phenotypic characteristics non-invasively by applying feature algorithms to medical imaging data.In this study of lung adenocarcinoma,we investigated the association between radiomic features and the tumor histologic invasiveness(pre-invasive lesions and invasive lesions).Furthermore,in order to predict the histologic invasiveness,we employed machine-learning methods and evaluated their prediction performance.Methods: We included all of the102 patients with lung adenocarcinoma(60 in pre-invasive group and 42 in invasive group)for this retrospective study.A total of 93 radiomic features were extracted from the segmented tumor volumes of pretreatment CT images.These radiomic features quantify tumor phenotypic characteristics on medical images using tumor shape,intensity statistics,and texture.A Mann-Whitney U-test and information gain algorithm were used to select radiomic features,and the support vector machine,naive Bayes and logistic regression classifier model was established and the ROC curve was used to evaluate their prediction performance.Results: Among 93 features were derived from each VOI,48 features were was selected by Mann-Whitney U-test and the information gain algorithm.The area under the receiver operating characteristic curve of support vector machine,Naive Baye's classifier and logistic regression classifier model were 0.822,0.848 and 0.874.Conclusion: The radiomic features from CT images can reflect the difference between preinvasive lesion and invasive lesion of p GGN lung adenocarcinoma less than or equal to 10 mm.Radiomic features capturing detailed information of the tumor phenotype can be used to identify tumor invasiveness.The classification model based on the radiomic features can improve the preoperative for the prediction of invasiveness of p GGN.