Analysis Of Characteristic Exhaled Breath Molecule In Patients With Lung Cancer Based On EESI-MS

Objective:The incidence of lung cancer is extremely high,and it is the leading cause of cancer death around the world.Early detection of lung cancer is limited since inexpensive,non-invasive and sufficiently sensitive and specific screening methods are not available.In this study,exhaled breath in healthy control and lung cancer patients were analyzed using EESI-MS respectively to preliminarily identify differentiated volatile organic compounds(VOCs)and analyze the molecular formulas in the exhaled breath of lung cancer patients.Models for distinguishing between lung cancer and healthy people(machine learning and pattern discrimination)were established to evaluate the diagnostic efficacy.Methods:135 lung cancer patients who were admitted to the respiratory department,oncology department or cardiothoracic surgery department in the first affiliated hospital of Nanchang University from July 2016 to September 2017 and 38 healthy controls who were admitted to physical examination department at the same time were enrolled.Exhaled breath of the patients in two groups were collected using the Tellar bag and the exhaled breath samples were detected by EESI-MS to obtain a first-order mass spectrum.Then diagnostic models were established to evaluate the diagnostic efficacy.The characteristics of VOCs in the exhaled breath of lung cancer patients were analyzed using different statistical methods and analysis softwares.Results:(1)The first-order mass spectrum shows that the exhaled components of lung cancer patients and healthy people are different(as judged by the mass-to-charge ratio),and the same exhaled breath composition also differs in the content of the two groups(signal intensity is different).(2)In this experiment,there was no statistical difference in smoking,age and gender between the lung cancer group and the healthy group.The sum of the contribution of PC1,PC2,and PC3 to the covariance of the exhaled breath in the two groups on the PCA was 88.39%,with statistically significant significance,and they were difficient.There are also differences between the exhaled breath of two groups on t-SNE.(3)The representative data of lung cancer group and healthy group were extracted,and three kinds of machine learning diagnosis models such as stochastic gradient descending classifier(SGD)model,logistic regression classifier model and random forest classifier model were established.The accuracy of identifying lung cancer of the three models was 97.83%,96.75% and 96.26% respectively,and the best model was the SGD model.(4)Through statistical analysis using the PCA and SPSS statistical software,wefound seven kinds of VOCs,including 2-acetylpyrrole,2-(dimethylamino),cyclohexanone,decane,carveol,22-ethylhexylacetate,2,3,5,4-thtrahydroxystilnene-2-O-D-glucoside and 2-(3-IODOPHENYL)PYRROLIDINE.which can distinguish between lung cancer patients and healthy people(one of them could not be found the Chinese name).Conclusion:(1)The composition and content of exhaled breath differed between lung cancer patients and healthy people,and the exhaled breath mainly contained VOCs.Lung cancer patients may contain seven characteristic VOCs such as 2-acetyl pyrrole,2-(dimethylamino)cyclohexanone,decane,carveol,22-ethylhexylacetate,2,3,5,4-thtrahydroxystilnene-2-O-D-glucoside and 2-(3-IODOPHENYL)PYRROLIDINE.(2)The high accuracy of identifying lung cancer of machine learning model makes it hopeful to use an early diagnosis model of lung cancer with high specificity and high sensitivity for our clinical work in the future.