基于呼气中特异性VOCs筛查的肺癌早期诊断与模型评估

Early Diagnosis of Lung Cancer Based on Breath Specific VOCs Screening and Model Assessment

采用气相色谱-质谱联用仪(GC-MS)检测了180例受试者的呼出气样品,包括79名肺癌患者和101名健康志愿者。每个受试者采集3个平行样品,以及1个室内空气样品。对所有呼气样品中检测的92种挥发性有机物(VOCs)进行定量分析。结合Mann-Whitney检验和正交偏最小二乘判别分析(OPLS-DA)模型筛选出10种肺癌患者呼气特异性VOCs,包括苯甲醛、顺式-2-丁烯、2-丁酮、萘、乙酸乙烯酯、乙烯、2,2,4-三甲基戊烷、3-甲基戊烷、己醛和2-甲基戊烷。利用统计学方法研究其在不同人群中的代谢差异和可能相关的代谢机制,通过建立机器学习模型验证候选标志物对疾病的诊断性能,结果显示,随机森林模型诊断的准确度、精准率、灵敏度和特异性分别为96.25%、96.21%、95.76%、96.67%,马修斯相关系数(MCC)为0.93,曲线下面积为0.96。上述10种化合物可作为肺癌患者的潜在呼气VOCs标志物,为肺癌的早期诊断提供了丰富的基础数据。

A gas chromatography-mass spectrometry(GC-MS)was adopted to detect the breath samples of 180 subjects in this paper,including 79 patients with lung cancer and 101 healthy controls.A total of 92 volatile organic compounds(VOCs)were detected.Furthermore,10 potential characteristic VOCs for lung cancer patients were screened out by orthogonal partial least-squares discrimination analysis(OPLS-DA)model combined with Mann-Whitney test,including benzaldehyde,cis-2-butene,2-butanone,naphthalene,vinyl acetate,ethylene,2,2,4-trimethylpentane,3-methylpentane,hexanal and 2-methylpentane.Thereafter,the metabolic differences in different populations and possible related metabolic mechanisms were analyzed using statistical method.Then,the diagnostic performance of candidate breath VOC biomarkers for lung cancer were verified by establishing machine learning models.The results showed that,the diagnostic accuracy,precision,sensitivity and specificity of random forests were 96.25%,96.21%,95.76%and 96.67%,respectively,and the Matthews correlation coefficient(MCC)was 0.93,the area under ROC curve was 0.96.All the 10 compounds could be taken as the potential breath VOC biomarkers for discriminating lung cancer and health people,which have supplied rich basic data for early diagnosis of lung cancer.