Exploring Quantitative Information On The Invasiveness Of Pulmonary Subsolid Nodules Based On Artificial Intelligence Software

Background and Purpose:Lung cancer is the leading cause of cancer death in both men and women^[1].Lung adenocarcinoma is the most common histological classification of lung cancer.With the gradual promotion and popularization of lung cancer screening using low-dose CT,more and more pulmonary nodules are detected,especially subsolid nodules（SSN）.Atypical adenomatous hyperplasia（AAH）,adenocarcinoma insitu（AIS）,minimally invasive adenocarcinoma（MIA）and invasive adenocarcinoma（IAC）can all manifest as SSN.Accurately distinguish between AAH/AIS/MIA and IAC,lepidic predominant adenocarcinoma（LPA）and non-LPA in IAC has important clinical value.The purpose of this study was to compare the differences in indolent lesions（including AAH,AIS,MIA）manifesting as SSN with IAC,LPA in IAC and non-LPA based on the quantitative information extracted by artificial intelligence software to obtain more sensitive predictions indicators and analyze their best predicted values.It provides a new basis for the differential diagnosis of indolent disease and IAC,LPA and non-LPA in IAC,and better guides the choice of clinical treatment plan.Materials and Methods:A total of 602 patients who were admitted to China-Japan Friendship Hospital of Jilin University from July 2020 to August 2021 and underwent preoperative chest CT（computed tomography）examination in our department and were confirmed to be AAH,AIS,MIA or IAC by postoperative pathology were collected.example.Two radiologists who were unaware of the clinical data and pathological results of the patients independently read the radiographs,and recorded general chest CT findings,including gender,age,lesion location,margins,borders,the presence or absence of vacuoles,and the presence or absence of pleural indentation.The CT maximum value,CT minimum value,CT average value,CT value variance,kurtosis,skewness,energy,average length and short diameter,3D length of SSN were obtained from CT image-assisted detection software（Hangzhou Shenrui Bolian Technology Co.,Ltd.）.diameter,compactness,sphericity,entropy.All specimens were pathologically classified according to the 2011 International Society for the Study of Lung Cancer（IASLC）/American Thoracic Society（ATS）/European Respiratory Society（ERS）classification of lung adenocarcinoma by two pathologists.SPSS 26.0 statistical software was used for statistical analysis of the data.The data were divided into indolent lesions group and IAC group,LPA group and non-LPA group（including acinar-predominant lung adenocarcinoma and papillary-predominant lung adenocarcinoma）.The t-test was used for normally distributed continuous variables and the Mann-Whitney U test for non-normally distributed continuous variables.Categorical variables were analyzed with Pearson’s chi-square test and Fisher’s exact test.Indolent lesions were matched 1:1 with IAC and LPA with non-LPA,respectively,by simple random sampling of the IAC group and the non-LPA group.Multivariate logistic regression analysis was used to evaluate independent predictors of IAC and non-LPA.The receiver operating characteristic curve（ROC curve）was drawn to determine the diagnostic sensitivity,specificity,accuracy and area under the ROC curve（AUC）of each parameter.The difference of each ROC curve was compared by Medcalc software.Results:In the analysis of the general clinical and imaging characteristics of the patients,there was no significant difference in gender,location,kurtosis,skewness and entropy between the indolent lesion group and the IAC group（P>0.05）.Differences in age,average lesion length and short diameter,3D long diameter,maximum cross-sectional area,surface area,volume,mass,CT maximum value,CT minimum value,CT average value,CT value variance,energy,compactness,and sphericity were statistically significant Significance（P<0.05）.Multivariate logistic regression analysis showed that quality（OR=1.008,95%CI 1.005-1.012,P<0.001）and CT mean（OR=1.004,95%CI 1.001-1.007,P=0.017）were independent predictors of IAC,The probability model for predicting SSN as IAC is:P=0.008X1+0.004X2-2.184,where X1 is the quality and X2 is the average CT value,and the prediction accuracy of this model is 77.7%.The ROC curve was drawn according to the logistic regression results.The area under the predicted probability curve（AUC）was 0.853,the AUC for the mass and CT mean values??were 0.853 and 0.708,respectively;the mass cutoff value was 124.3 mg,the diagnostic sensitivity was 0.865,and the specificity was 0.723;the cutoff value for the CT mean value was-532.69HU,the diagnostic sensitivity was 0.745,and the specificity was 0.617.By comparing the ROC curves of the three indicators by Medcalc software,the difference between the predicted probability and quality was not statistically significant（P=0.1829）;CT mean and prediction The ROC curves of probability,CT mean and quality were statistically different（P<0.001）.There were no significant differences in gender,age,location,CT minimum value,kurtosis,skewness,and entropy between the LPA group and the non-LPA group（P>0.05）.There were statistically significant differences in area,surface area,volume,mass,CT maximum value,CT average value,CT value variance,energy,compactness and sphericity（P<0.05）.Multivariate logistic regression analysis showed that mean CT（OR=1.004,95%CI1.002-1.007,P<0.001）was an independent predictor of non-LPA IAC.The ROC curve was drawn according to the logistic regression results.The AUC of the mean CT was 0.727,the cutoff value was-454.76HU,the diagnostic sensitivity was 0.645,and the specificity was 0.785.ConclusionQuantitative information based on artificial intelligence software can be used to diagnose SSN aggressiveness.For indolent lesions and IAC,quality is the most valuable independent predictor;for LPA and non-LPA in IAC,CT mean is the most valuable independent predictor.