CT Radiomics Model For The Evaluation On The Invasion Of The Lung Adenocarcinoma As Glass Ground Nodules

Objective:To investigate the ability of CT radiomics model for diagnosing invasion of the lung adenocarcinoma as ground-glass nodules(GGNs).Materials and Methods:Two hundred and ninety cases of the lung adenocarcinoma asground glass nodules confirmed by pathology between January 2017 and April 2018 were collected in this study,in accordance with the complete clinical,thin layer imagesimaging and pathological data.There were 73 males and 217 females,aged 26-80 years,with an averageage of 56.61±10.06 years.There were 88 cases of preinvasive adenocarcinoma,91 cases of minimallyinvasive adenocarcinoma(MIA)and 111 cases of invasive adenocarcinoma(IAC).According to stratified random sampling,all cases were divided into training cohort(n =204)and validationcohort(n =86).The tumor regions of interest(ROIs)were segmented manually by by two experienced radiologistsusing the Artificial Intelligent Kit(A.K.,GE Healthcare)software,and the imaging signs of the tumor were analyzed and recorded.The least absolute shrinkage and selection operator(LASSO)algorithm was used to select the most predictive features.And the radiomics signatures were formed by linear combinations of the selected features and the weighting coefficients.The final radiomics signatureswere further used to construct the radiomics models.Radiomics model cbility was evaluated in the training cohort and validated in the validation cohort.The area under the receiver operating characteristic curve(AUC)were used to evaluate the predictive effectiveness of radiomics models.The imaging signs recorded in terms include:morphology(spherical),size(maximum mean diameter),mean CT value,margin condition(lobular sign,burr sign),internal condition(vacuole sign,air bronchus sign),and adjacent structure condition(vascular cluster sign,pleural concave sign).Independent t-test or Mann-Whitney U test was used for continuous variables and Fisher’s exact test or chi-square test was used for categorical variables.There was a statistical significance when P was less than 0.05.Results:A total of 396 radiomics features were extracted and 17 features were found to be the most important differential diagnostic features through dimensionality reduction.By multivariate logistic regression analysis,the CT-based radiomics modelwas constructed with good predictiveeffect.The CT-based radiomics model yielded higher AUC in training cohort in differentiating preinvasive adenocarcinomas(0.93),MIAs(0.85)and IACs(0.90),and reached an AUC of 0.95,0.77,0.80 in validating cohort,respectively.Radiomics model also showed favorable discrimination with an accuracy of 73.35% and72.09% in the training cohort and the validation cohort,respectively.Encouragingly,the radiomics model showed greatest performance in differentiating preinvasive adenocarcinomas with a sensitivity of 84.62% and 96.97%,a specificity of 85.71% and75.47%in the training and validation cohort,respectively.As the table shows that the radiomics model had a greatest ability to differentiate preinvasive adenocarcinomas from MIAs-IACs group and IACs from pre-MIAs group,with a sensitivity of 84.62% and96.97%,a specificity of 85.71% and 75.47%in the training and validation cohor.However,the radiomics model hada relatively low ability to differentiate MIAs from pre-IACs groupwith a sensitivity of 60.94%,48.15%,a specificity of 84.29%,91.53%in the training and validation cohor.The present study also analyzed clinical and traditional imaging signs.The statistics showed that there were no significant differences in the patient’s sex,age,smoking historyfor diagnosing invasion of the lung adenocarcinoma.There werestatistically significant differences in tumor size,mean CT value,lobular sign,burr sign,and vacuolesignin the three groups of lesions(P<0.05).Morphology(spherical),air bronchus sign,vascular cluster sign,pleural indentation signhad no significant difference(P>0.05).Conclusion: As a non-invasive preoperative prediction tool,the CT radiomics model we have built has a good predictive performance to differentiatepreinvasiveadenocarcinomas,MIAs and IACs,especially in differentiating preinvasive adenocarcinomas.