Clinical Study In Predicting The Infiltration Degree Of Early-stage Lung Adenocarcinoma Appearing As Pulmonary Nodules Based By Artificial Intelligence-assisted Diagnosis System

Objective:To explore the efficacy of artificial intelligence-assisted diagnosis system in qualitatively diagnosing and quantitatively predicting the infiltration degree of early-stage lung adenocarcinoma of pulmonary nodules,and to further analyze the correlation between quantitative parameters of artificial intelligence-assisted diagnosis system and Ki-67.Methods:The clinical data of 247 patients with early-stage lung adenocarcinoma(267 pulmonary nodules)admitted to the Department of Thoracic Surgery,Lanzhou University Second Hospital from January 1,2015 to December 31,2022 were collected.The DICOM images of preoperative chest CT were imported into the artificial intelligence-assisted diagnosis system to complete automatic detection and intelligent analysis.Firstly,record and analyze whether CT qualitative and quantitative indicators can identify the degree of pulmonary nodule infiltration,and the confusion matrix and receiver operating characteristic curve(ROC)were further drawn to evaluate the efficacy of artificial intelligence-assisted diagnosis system in qualitative diagnosis of pulmonary nodule infiltration.Secondly,we analyzed the data of patients with pulmonary nodules whose Ki-67 can be obtained by immunohistochemical staining of postoperative specimens,and drew ROC curve to evaluate the efficacy of quantitative parameters of artificial intelligence-assisted diagnosis system and Ki-67 in predicting the degree of pulmonary nodule infiltration,and calculated the best cut-off value of Ki-67.Finally,to explore the correlation between different Ki-67 expression levels and quantitative parameters of artificial intelligence-assisted diagnosis system.Result:The diameter,maximum CT value,average CT value,central CT value,rate of invasive adenocarcinoma(IAC rate),pleural indentation sign,lobulation sign and spiculation sign of pulmonary nodules with different degrees of infiltration were statistically significant(P<0.05).With the increase of infiltration degree and density of pulmonary nodules,the proportion of dominant signs in each group increased gradually.In the differential diagnosis of precursor gland lesions(PGL),microinvasive adenocarcinoma(MIA)and invasive adenocarcinoma(IAC),the F1 score,the area under the curve(AUC),accuracy,recall rate and precision of artificial intelligence-assisted diagnosis system were 65.21%,0.795,72.66%,70.98%and 63.49%,respectively.In the differential diagnosis of PGL and MIA+IAC,the AUC value of artificial intelligence-assisted diagnosis system was 0.860,and the sensitivity and specificity were 90.63%and 81.40%,respectively.In the differential diagnosis of PGL+MIA and IAC,the AUC value of artificial intelligence-assisted diagnosis system was 0.825,the sensitivity and specificity were 73.66%and 91.36%,respectively.A total of 215 pulmonary nodules(58 in PGL+MIA group and 157 in IAC group)with Ki-67 available by immunohistochemical staining of postoperative specimens were selected.Compared with the PGL+MIA group,the IAC group had longer diameter and larger volume,higher IAC rate,maximum CT value,minimum CT value,average CT value and central CT value,and higher detection rates of pleural indentation sign,lobulation sign and spiculation sign(P<0.001).The image features with statistical significance in univariate analysis were further included in binary logistic regression analysis.The results showed that diameter and average CT value were independent risk factors for IAC.The results of ROC curve analysis showed that the IAC rate had the best performance in differentiating PGL+MIA group from IAC group.The AUC value was 0.908,the best cut-off value was 44.50%,and the sensitivity and specificity were82.17%and 87.93%,respectively.The best cut-off value of Ki-67 was 6%,which was divided into≤6%group and>6%group.The above imaging features were significantly different between different Ki-67 groups(P<0.001),and correlation analysis showed that Ki-67 had the highest correlation with average CT values(r_s=0.553).Binary logistic regression analysis showed that average CT value(P=0.044)was an independent risk factor for Ki-67 expression.Conclusions:The artificial intelligence-assisted diagnosis system predicts the degree of infiltration of pulmonary nodules through preoperative CT images,and verifies each other with postoperative pathological diagnosis.It can stratify the infiltration of pulmonary nodules from both quantitative and qualitative aspects,and has good diagnostic efficiency.Preoperative CT imaging parameters of pulmonary nodules were correlated with Ki-67,and the average CT value was moderately correlated with Ki-67 expression.It is believed that with the continuous improvement of relevant training data and the improvement of deep learning network algorithm,the artificial intelligence-assisted diagnosis system may be able to analyze and predict the infiltration and proliferation of pulmonary nodules more comprehensively in the future,so as to provide valuable reference for individualized diagnosis and treatment of patients.