Study on Repeatability, Normalization and Feature Selection of Medical Electronic Nose for Lung Cancer Diagnosis

Abstract

In this paper, we report a medical electronic nose for lung cancer diagnosis used for breath analysis. We use the designed alveolar gas collector to collect the alveolar gas of subjects for the electronic nose experiment. The electronic nose system has completed chemical and exhaled breath tests. The results of chemical tests show the good repeatability of the developed medical electronic nose for lung cancer diagnosis in this paper, and the relative standard deviation (RSD) values of five feature classes are calculated, respectively. In the exhaled breath test, the test data are processed with six normalization methods. It is found that the data classification effect of the data after the quantile normalization is better than that after the other five normalizations. Then, the five feature classes are classified separately, and it is found that the feature class with low RSD value obtained from chemical tests has better classification effect, which provides a new way for feature selection.