Study On Correlation Between The Imaging Manifestations And Pathology Of Ground Glass Pulmonary Nodules With Artificial Intelligence

Objective:To explore the correlation between CT parameters and pathological types of pulmonary ground glass nodules(GGN)by using artificial intelligence aided diagnosis system,so as to provide imaging basis for early detection,early diagnosis and optimal treatment of GGN.Methods:103 ground-glass nodules of lung adenocarcinoma confirmed by surgery and pathology in our hospital from January 2019 to January 2021 were included and analyzed with artificial intelligence software workstation.103 chest CT quantitative parameters recorded after analysis included nodule volume,mean CT value,kurtosis,maximum surface area,surface area,3D length diameter,compactness,sphericity,and entropy.Qualitative parameters included lobulation sign,burrs,edge clarity and density uniformity.103 GGNs were divided into non-invasive group(AAH,AIS)and invasive group(MIA,IAC)according to the standard of invasive lung adenocarcinoma.According to the WHO new classification standard for lung adenocarcinoma in 2015,103 GGNs were classified as follows:AAH,AIS,MIA and IAC,using SPSS and Med Calc statistical analysis software,the difference between the two groups to have statistical significance of quantitative parameters for the receiver-operating characteristic curve(Receiver oPerating characteristic curves,the ROC)analysis,evaluate its diagnostic GGN invasive ability,at the same time,according to the most about an index(Youden’s index,YI)calculate the quantitative parameters of the best diagnostic threshold,the area under the curve,sensitivity and specific degrees,P<0.05 was considered statistically significant,while P>0.05 showed no statistically significant difference.Results:1)3D length diameter,average CT value,volume,maximum area and surface area were significantly different between invasive and non-invasive groups,as well as between AAH and AIS,AIS and MIA,and MIA and IAC(P<0.05).The CT quantitative parameters of nodules,such as compactness,sphericity,entropy and kurtosis,had no statistical differences among all groups(P>0.05).2)ROC curve analysis:①The threshold value of 3D long diameter in non-invasive and invasive lesions was 11.82mm,the sensitivity was 70%,and the specificity was 72.7%;The threshold of AAH and AIS lesions was 7.78mm,the sensitivity was 96.9%,and the specificity was 75%.The threshold of AIS and MIA lesions was 11.51mm,the sensitivity was 92.3%,and the specificity was 73%.The threshold of MIA and IAC changes was 12.75mm,the sensitivity was 70%,and the specificity was 92.3%.②The threshold of mean CT value in non-invasive and invasive lesions was-413.7Hu,the sensitivity was 87.1%,and the specificity was 60.6%.The threshold of AAH and AIS lesions was-592.9HU,the sensitivity was 60.3%,and the specificity was 94.3%.The threshold of AIS and MIA lesions was-488.7HU,the sensitivity was 84.61%,and the specificity was 65.1%.The threshold of MIA and IAC changes was-418.7HU,the sensitivity was 98.2%,and the specificity was 76.9%.③The threshold value of the volume in non-invasive and invasive lesions was 590mm3,the sensitivity was 71.4%,and the specificity was 66.7%.The threshold of AAH and AIS lesions was 335mm3,the sensitivity was 82.54%,and the specificity was 87.50%.The threshold of AIS and MIA lesions was 757mm3,the sensitivity was 84.62%,and the specificity was 77.78%.The threshold value of MIA and IAC changes was 1090mm3,the sensitivity was 65.0%,and the specificity was 92.3%.④The threshold of the maximum surface area in non-invasive and invasive lesions was 114.6mm2,the sensitivity was 85.7%,and the specificity was 54.6%.The threshold of AAH and AIS lesions was 45.18mm2,the sensitivity was 81.0%,and the specificity was 87.5%.The threshold of AIS and MIA lesions was 73.38mm2,the sensitivity was 98.2%,and the specificity was 60.3%.The threshold value of MIA and IAC changes was 116.2mm2,the sensitivity was 70.0%,and the specificity was 85.71%.⑤The threshold value of non-invasive surface area and invasive lesions was 363.69mm2,the sensitivity was 70%,and the specificity was 72.7%.The threshold of AAH and AIS lesions was 222.84mm2,the sensitivity was 87.30%,and the specificity was 75.0%.The threshold of AIS and MIA lesions was 363.69mm2,the sensitivity was 92.31%,and the specificity was 66.7%.The threshold value of MIA and IAC changes was 547.5mm2,the sensitivity was 70.0%,and the specificity was 84.6%.3)The nodular edge signs,such as lobulation sign,burr sign,edge clarity and density uniformity,were significantly different between the invasive and non-invasive groups(P<0.05).There were significant differences between AIS and MIA in nodular edge signs,such as lesion lobulation sign,burr sign,edge clarity and density uniformity(P<0.05).There were significant differences in nodular density uniformity between AAH and AIS,AIS and MIA,and MIA and IAC(P<0.05).Conclusion:1.The application of artificial intelligence in CT quantitative analysis(3D length,CT mean value,volume,maximum area,surface area)has diagnostic significance for the prediction of invasive and non-invasive GGN and different pathological types of GGN.2.The marginal signs of GGN can provide reference for the diagnosis of pathological type and infiltration of GGN.