| Objective:To explore the clinical value of artificial intelligence quantitative parameters of lung nodules to assess different infiltration degrees of early lung adenocarcinoma with GGN(pGGN and mGGN)on CT.Methods:A retrospective collection of patients in the Thoracic Surgery or Elderly Thoracic Surgery Department of the First Affiliated Hospital of Kunming Medical University from July 2018 to July 2020.GGN was detected by HRCT of the chest,and sublobular/segment/lobectomy was implemented,the pathology all suggest early lung adenocarcinoma.Finally,114 pGGN patients(AIS:29 cases,MIA:39 cases,IAC:46 cases)were enrolled,female/male:79/35,with an average age of 51.7±10.9 years.102 cases of mGGN(MIA:49 cases,IAC:53 cases)were enrolled,female/male:71/31,average age 54.03±10.2 years.Transfer the patient’s preoperative 1mm thin layer of lung window into the AI lung nodule detection system to obtain relevant quantitative parameters:quality/solid component,CT maximum,CT minimum,CT average,CT value variance,kurtosis,Skewness,energy,maximum area,surface area,3D long diameter,compactness,sphericity,entropy.According to the AIS,MIA,IAC related parameters of pGGN,single-factor ordered multi-class logistic regression analysis was performed,and variables with P<0.2 were selected for multicollinearity diagnosis and factor analysis.The effective variables and factors in the above results were incorporated into the ordered multi-class Logistic regression model to analyze the independent risk factors related to the degree of pGGN early lung adenocarcinoma infiltration.Taking pGGN’s AIS and MIA as the research objects,a binary logistic regression analysis was performed,and the indexes with P<0.05 were selected for ROC curve analysis,and the corresponding area under the curve,threshold,sensitivity,and specificity were analyzed.Similarly,using pGGN’s MIA and IAC as the research objects,we conducted binary logistic regression analysis,and selected P<0.05 indicators for ROC curve analysis,and analyzed the corresponding area under the curve,threshold,sensitivity,and specificity.In addition,with mGGN early lung adenocarcinoma as the research object,binary Logistic regression analysis was performed according to MIA and IAC,and P<0.05 indicators were selected for ROC curve analysis,and the corresponding area under the curve,threshold,sensitivity,and specificity were analyzed.Results:1、The difference 3D major diameter,surface area,maximum area,mass,CT maximum,CT value variance,compactness,sphericity,and energy is statistically significant in the three groups of pGGN’ s AIS,MIA,IAC(P<0.05).Among them,mass,3D long diameter,and largest area are statistically different in AIS,MIA,and IAC(P<0.05).At the same time,surface area,compactness,and sphericity are statistically different in the AIS/IAC group and MIA/IAC group(P<0.05),while the maximum CT value,variance of CT value,and energy are only statistically different in the AIS/IAC group(P<0.05).In addition,whether smoking,gender,age,CT minimum,and CT average were in the three groups of pGGN AIS,MIA,IAC data distribution,the difference was not statistically significant(P>0.05).In addition,for mGGN,the differences in 3D major diameter,surface area,maximum area,CT maximum,kurtosis,and energy were statistically significant(P<0.05).2、The orderly single factor Logistic analysis of AIS,MIA,IAC of pGGN showed:quality,CT maximum,CT minimum,CT value variance,skewness,maximum area,surface area,3D long diameter,compactness,sphericity was statistically significant(P<0.05).The diagnosis of multicollinearity suggests that there is a collinearity problem in the above 10 variables.The principal component analysis in factor analysis is further carried out,and finally two effective common factors F1 and F2 are extracted.The results of principal component-ordered multivariate logistic analysis showed that common factor F1 odds ratio(OR=8.611,P=0.001)and common factor F2 odds ratio(OR=1.833,P=0.012)were early lung adenocarcinoma pGGN Independent risk factors for invasiveness.3、A one-way binary logistic regression analysis was performed on the AIS and MIA of pGGN,and it was found that mass,energy,maximum area,3D long diameter,and surface area were statistically different between the two groups(P<0.05).The AUC,threshold,sensitivity,and specificity of the above indicators are as follows:mass(0.734,198.5mg,0.862,0.538),energy(0.728,72.960×1014,0.862,0.487),maximum area(0.698,72.960mm~2,0.862,0.487)),3D long diameter(0.696,11.234mm,0.828,0.513),surface area(0.691,282.02mm~2,0.897,0.538).In addition,the combination of the above indicators can slightly increase the overall diagnostic predictive value corresponding to the AUC,sensitivity,and specificity scores(0.760,0.795,0.621).4、A single-factor binary Logistic regression analysis was performed on the MIA and IAC of pGGN,and it was found that the surface area,mass,3D long diameter,maximum area,CT maximum,compactness,and sphericity were statistically different between the two groups(P<0.05).Surface area,mass,3D major diameter,maximum area,and CT maximum are risk factors for increased infiltration.The corresponding AUC,threshold,sensitivity,and specificity are as follows:surface area(0.819,415 mm~2,0.897,0.674),mass(0.813,342mg,0.897,0.609),3D long diameter(0.779,11.965mm,0.641,0.783),maximum area(0.761,82.735 mm~2,0.718,0.696),CT maximum(0.658,-122HU,0.872,0.543).In addition,comprehensive surface area,mass,3D long diameter,maximum area,and maximum CT value can slightly improve the overall predictive value of IAC,corresponding to AUC,sensitivity,and specificity scores(0.839,0.739,0.846).The compactness and sphericity are protective factors for increased infiltration,and the corresponding AUC,threshold,sensitivity,and specificity are as follows:compactness(0.688,0.825,0.462,0.783),sphericity(0.686,0.885,0.179,0.500).5、A single-factor binary logistic regression analysis was performed on the MIA and IAC of mGGN,and it was found that the solid component,the maximum CT value,the maximum area,the surface area,and the 3D long diameter were statistically different between the two groups(P<0.05).The above indicators were further analyzed by ROC curve.In addition to the solid component P>0.05,the surface area,3D major diameter,maximum area,and CT maximum value were all P<0.05,and the corresponding AUC,threshold,sensitivity,and specificity were respectively the surface area(0.773,376.55mm~2,0.633,0.830),3D major diameter(0.770,14.125mm,0.673,0.774),maximum area(0.758,96.61mm~2,0.673,0.755),CT maximum(0.689,141.5HU,0.857,0.453).In addition,joint index can slightly improve the diagnostic predictive value of overall mGGN,corresponding to AUC,sensitivity,and specificity scores(0.825,0.792,0.735).Conclusions:1.The quantitative parameters of AI nodules based on deep learning help to identify the pathological subtypes of GGN with different degrees of infiltration.2.Different quantitative parameters of AI nodules have different diagnostic efficacy in assessing the degree of pGGN or mGGN infiltration,while the three-dimensional quantitative parameters of nodules(3D long diameter,maximum area,surface area)play a major role. |
PDF Full Text Request