The Value Of CT-Radiomics In Judging The Invasiveness Of Pulmonary Heterogeneous Ground Glass Nodules

PART One A comparative study between CT signs and pathological features in judging invasion of pulmonary HGGNsPurpose:To explore the relationship between the CT signs of pulmonary HGGNs and the pathological features of invasiveness of the lesions,and to improve the accuracy of CT diagnosis of pulmonary HGGNsMaterials and Methods:326 pulmonary nodules(smaller than 30mm)confirmed by pathology from 2016 to 2018 were collected retrospectively.A total of 125 persistent HGGNs(?3months)were enrolled,which were confirmed by two senior radiologist,including 58 cases of AIS,42 cases of MIA,and 25 cases of IAC.CT signs include:the size of HGGNs,mean CT values,the shape of heterogeneous opacities in the lesions,bubblelike lucency,border definition and spiculation,lobulation,vascular convergence or attachment,bronchial structures associated with the lesions,pleural retractionResult:There were significant differences in lesion size and mean CT values between AIS?MIA and IAC(P<0.05).The critical values were 9.5 mm and-535 HU,respectively.The types of heterogeneous opacities in the lesions include--"circular","circular with punctate" and "branch" images,which were significantly different between AIS and MIA(P<0.05).The signs of bubblelike lucency,spiculation and lobulation were significantly different between MIA and IAC(P<0.05).There were no significant differences in gender,adjacent bronchial structures,vascular convergence or attachment,pleural retraction changes,"curved moon" shadow,"dot-like" shadow between AIS,MIA and IAC(P>0.05).Conclusion:Comprehensive analysis of the thin-slice CT signs of persistent HGGNs lesions can help identify pulmonary adenocarcinomas as AIS,MIA or IACPART Two Application of CT-Radiomic features for predicting histological invasiveness in pulmonary HGGNsOBJECTIVE:To evaluate the value of radiomics signature based on thin-sliced CT of pulmonary HGGNs for predicting the invasiveness of lesions.Materials and Methods:A retrospective study of 125 consecutive HGGNs(?3 months)was performed,and all cases were pathological confirmed.All the images of cases were non-contrast thin-sliced CT that were reconstructed with a standard kernel of a slice thickness of 1.5 mm.After using ITK-Snap software to manually delineate the lesion range,the lesion data was input into A.K.software(AnalysisKit 3.2.2,GE Healthcare)for processing.All cases were randomly selected as a training cohort and a test cohort at 7:3 The characteristics of the training cohort data were analyzed by Spearman correlation analysis,LASSO regression analysis and random forest analysis to construct the radiomics signature.ROC analysis in test cohort is used to evaluate the predictive power of radiomics signature.RESULTS:Firstly,radiomics score calculation formula was formed to identify the AIS from the invasive lesions(MIA+IAC).The AUC in test cohort was 0.99.The sensitivity,specificity,accuracy were 0.96,0.86,and 0.94,respectively.Secondly,the radiomics signature is established to identify AIS,MIA,and IAC.The AUCs in test cohort were 0.91,0.84,0.97,respectively.Conclusion:Radiomics signature to predict pre-invasive/invasive lesions or to distinguish AIS,MIA from IAC has high sensitivity,specificity and accuracy,which provides a reliable basis for the diagnosis and treatment.