Background: Lung cancer is one of the most common malignant tumors worldwide. Currently, effective screening methods for early lung cancer are still scarce. Breath analysis provides a promising method for the pre-screening or early screening of lung cancer. Isoprene is a potential and important breath biomarker of lung cancer. Material and Methods: To investigate the clinical value of isoprene for diagnosing lung cancer patients, a cavity ringdown spectroscopy (CRDS) based near-real time, sensitive analysis method of breath isoprene is developed in our lab. In this paper, 92 breath samples from lung cancer patients, 17 breath samples from patients with benign lesions, and 107 breath samples from healthy people were collected. Results: Research indicates that breath isoprene concentration is significantly higher in healthy individuals (221.3±122.2ppbv) than in patients with lung cancer (112.0±36.6ppbv) and benign lung lesions (127.9±41.2ppbv). The result of Receiver Operating Characteristic (ROC) curve suggests that the concentration of isoprene is meaningful for the diagnosis of lung cancer (AUC=0.822, sensitivity=63.6%, specificity=90.2%, P<0.01). Conclusion: This study demonstrates that the CRDS breath isoprene analysis system can effectively analyze a large sample of human breath isoprene, and preliminarily confirms the use of breath isoprene as a biomarker for lung diseases.Background: Lung cancer is one of the most common malignant tumors worldwide. Currently, effective screening methods for early lung cancer are still scarce. Breath analysis provides a promising method for the pre-screening or early screening of lung cancer. Isoprene is a potential and important breath biomarker of lung cancer. Material and Methods: To investigate the clinical value of isoprene for diagnosing lung cancer patients, a cavity ringdown spectroscopy (CRDS) based near-real time, sensitive analysis method of breath isoprene is developed in our lab. In this paper, 92 breath samples from lung cancer patients, 17 breath samples from patients with benign lesions, and 107 breath samples from healthy people were collected. Results: Research indicates that breath isoprene concentration is significantly higher in healthy individuals (221.3±122.2ppbv) than in patients with lung cancer (112.0±36.6ppbv) and benign lung lesions (127.9±41.2ppbv). The result of Receiver Operating Characteristic (ROC) curve suggests that the concentration of isoprene is meaningful for the diagnosis of lung cancer (AUC=0.822, sensitivity=63.6%, specificity=90.2%, P<0.01). Conclusion: This study demonstrates that the CRDS breath isoprene analysis system can effectively analyze a large sample of human breath isoprene, and preliminarily confirms the use of breath isoprene as a biomarker for lung diseases.

提出一种将卷积神经网络(CNN)学习特征与传统影像学特征结合的肺结节良恶性鉴别方法。首先,从电子计算机断层扫描(CT)图像中分割出肺结节区域,并使用传统机器学习方法提取结节区域的影像学特征;然后,使用截取的肺结节训练3D-Inception-ResNet模型,提取网络学习的CNN特征,组合两类特征,并利用随机森林(RF)模型进行特征选择;最后,采用支持向量机(SVM)、RF等传统分类器对肺结节进行良恶性鉴别诊断。使用LIDC-IDRI数据库中的1036个肺结节进行实验验证,最终所提方法的分类准确率、敏感度、特异度及接受者操作特性曲线(ROC)下面积(AUC)分别达94.98%、90.02%、97.03%及97.43%。实验结果表明,所提方法能准确地判别肺结节的良恶性,并优于大部分主流方法。