Application Research On Differential Diagnosis Of Benign And Malignant Pulmonary Nodules And Pathological Classification Of Lung Adenocarcinoma Based On Artificial Intelligence Aided Diagnosis System

Objective:To investigate the performance of artificial intelligence assisted diagnosis system in the differential diagnosis of benign and malignant pulmonary nodules and whether the addition of serum tumor markers is helpful to the differential diagnosis of benign and malignant pulmonary nodules,and to further explore the diagnostic performance of artificial intelligence assisted diagnosis system in the pathological classification of lung adenocarcinoma in subsolid nodules（SNNs）.Methods:The clinical data of 404 patients with pulmonary nodules（419pulmonary nodules）,diagnosed by pathology in the Second Hospital of Lanzhou University from June 2014 to December 2021,were retrospectively analyzed.The original DICOM format file of lung window of preoperative chest CT was imported into artificial intelligence aided diagnosis system(Surgipro^TMsystem).First,the patient data were recorded and analyzed,and the receiver operating characteristic（ROC）curve was used to analyze the relationship between the malignancy risk value of the SurgiPro^TMsystem and the benign and malignant pulmonary nodules,and to calculate the best cut-off value of the malignancy risk value.To further evaluate the diagnostic performance of SurgiPro^TMsystem in differentiating benign and malignant pulmonary nodules when the malignancy risk value was set to different cut-off values.Secondly,the above mentioned patients whose serum tumor markers carcinoembryonic antigen（CEA）and cytokeratin 19 fragment（CY211）were detected before surgery were selected.The patient data were recorded and analyzed,and the prediction probability of each combined model was calculated by binary logistic regression analysis,and the corresponding ROC curves were drew to further analyze the value of CEA and CY211alone or in combination and its combination with SurgiPro^TMsystem in the identification of benign and malignant pulmonary nodules.Finally,the above mentioned lung adenocarcinoma patients with SSNs were selected.The patient data were recorded and analyzed,and the diagnostic performance of the SurgiPro^TMsystem for pathological typing of lung adenocarcinoma in SNNs were evaluated.Result:1.In the experiment of SurgiPro^TMsystem in differentiating benign and malignant pulmonary nodules,404 patients were enrolled,including 187 males（46.3%）and 217 females（53.7%）,aged 12～79（51.8±11.5）years old,with a total of 419 lungs Nodules（257 malignant and 162 benign）.Compared with benign patients,malignant patients were older,more common in women（P<0.05）,and the nodules were smaller in size,lower in mean density,and had significantly higher malignancy risk value（P<0.05）.The area under the curve（AUC）of malignant risk value for predicting benign and malignant pulmonary nodules was 0.841（P<0.001）,and the best cut-off value was77.5%.When the malignancy risk value was respectively cut at 60%and 77.5%,the AUC of SurgiPro^TMsystem for differentiating benign and malignant pulmonary nodules were（0.700 vs.0.773,P<0.001）;The sensitivity,specificity,positive likelihood ratio and negative likelihood ratio were 91.83%vs.85.99%,48.15%vs.68.52%,1.77 vs.2.73,0.17 vs.0.20,respectively;The false positive rates were（51.85%vs.31.48%）and decreased by 20.37%;The false negative rates were（8.17%vs.14.01%）and increased by 5.84%.2.In the experiment of SurgiPro^TMsystem combined with CEA and CY211 to differentiate benign and malignant pulmonary nodules,a total of 352 pulmonary nodules（including 221 malignant and 131 benign）were included.Compared with benign patients,malignant patients were older and more common in females（P<0.05）,and the nodules were smaller in size,lower in mean density,and had significantly higher malignancy risk value and higher CEA and CY211（P<0.05）.The AUC of CEA combined with CY211,CEA and CY211 in predicting benign and malignant pulmonary nodules were 0.635（P<0.001）,0.609（P=0.001）and 0.563（P=0.047）,respectively.And there was significant difference in AUC between CEA combined with CY211 vs.CY211（P<0.05）,but there was no significant difference in AUC between CEA combined with CY211 vs.CEA（P>0.05）;The AUC of malignant risk value,malignant risk value combined with CEA,malignant risk value combined with CY211 and the three combined to predict benign and malignant pulmonary nodules were 0.859（P<0.001）,0.863（P<0.001）,0.859（P<0.001）and 0.864（P<0.001）,respectively.But there was no significant difference between the AUC of the combined models and malignant risk value（P>0.05）.3.After testing,the malignancy risk value of adenocarcinoma,other malignant tumors and benign lesions were distributed around 96.0%,80.0%and 62.5%,respectively.114 lung adenocarcinoma manifested as SNNs were further collected,including 19 preinvasive lesions（PIL）,20 minimally invasive adenocarcinoma（MIA）,and 75 invasive adenocarcinoma（IAC）.From PIL to MIA to IAC,the types of pulmonary nodules from pure ground glass nodules（p GGNs）to SNNs to solid nodules（SNS）were the main types,the long diameter and volume gradually increased,and the mean density gradually increased（P<0.05）,but there was no significant difference in the distribution of malignant risk value（P>0.05）;In the distribution of long diameter,there was no significant difference between PIL and MIA（P>0.05）,but there was significant difference between PIL or MIA and IAC（P<0.05）.The AUC of long diameter,volume and mean density in differentiate PIL-MIA and IAC were 0.816（P<0.001）,0.791（P<0.001）and 0.772（P<0.001）,respectively.The best cut-off values were 11.43mm,793.00mm~3 and-558.25HU,respectively.The sensitivity was 86.08%,69.62%and 77.22%,respectively.And the specificity was 83.33%,75.44%and 77.19%respectively.Conclusions:The SurgiPro^TMsystem has strong performance in differentiating benign and malignant pulmonary nodules.In the process of use,we should actively search for the best cut-off value of malignant risk value（here is 77.5%）,and at the same time,we should weigh the occurrence of false positive and false negative nodules.Serum tumor markers CEA and CY211 alone or in combination have weak performance in differentiating benign and malignant pulmonary nodules,and are not helpful for SurgiPro^TMsystem to differentiate benign and malignant pulmonary nodules.In addition,the malignancy risk value of the SurgiPro^TMsystem can not differentiate the invasive type of lung adenocarcinoma manifesting as SNNs,but its quantitative parameters such as length,volume and mean density are helpful to differentiate.When the long diameter,volume and average density are respectively greater than 11.43mm,793.00mm~3 and-558.25HU,the pathological classification of SNNs is more likely to be IAC.In short,with the maturity of technology and the improvement of performance,the SurgiPro^TMsystem is expected to become an auxiliary tool for thoracic surgeons to judge the benign and malignant and invasive degree of pulmonary nodules before operation and guide the treatment.