Prediction of Pulmonary Diseases With Electronic Nose Using SVM and XGBoost

Abstract

The volatile organic compounds (VOC) present in the exhaled breath can be used as the biomarkers of certain diseases especially pulmonary diseases. A system and device are required for the diagnosis of these diseases that can be applied easily, non-invasive, produce high accuracy results, and minimal side effects as possible. This work seeks to generate biomarkers from non-invasive and accessible breath samples that facilitate diagnostic strategies. The objective of this study is to establish breath fingerprints in human exhaled breath for the timely diagnosis of lung cancer, chronic obstructive pulmonary disease (COPD), and asthma through the use of metabolomics tools. An electronic nose (e-nose) system is developed for the analysis of exhaled breath, which was applied to detect and classify a set of exhaled breath samples from healthy people and patients with lung cancer, COPD, and asthma. Breath samples of 218 people, including 48 lung cancer patients, 52 COPD patients, 55 asthma Patients, and 63 healthy controls were evaluated. To evaluate the performance in discriminating patients from healthy controls, eight different machine learning models were designed. The KPCA-XGBoost model attained good results with accuracy, sensitivity, and specificity of 91.74%, 90.57%, and 92.65% respectively for lung cancer prediction; 89.84%, 88.14%, and 91.30% respectively for COPD prediction, and 70.66%, 68.75%, and 72.41% respectively for asthma prediction.