Study Of Medical Electronic Nose For The Detection Of Bacteria In Wound Infection

When used for wound infection detection, E-nose possess the characteristics of rapid detection non-invasive?efficient and no high requirements on the environment, and have wide application prospect. It is of great importance in clinical practice that designing of medical E-nose system realizes the detection of pathogenic bacteria in wound infection by detecting metabolites odor of common wound infection bacteria during growth.This piece aims to construct an E-nose odor collection platform which enable information collection of bacterial metabolites?data processing and analysis, achieving rapid screening of patients with wound infection of pathogenic bacteria. Simultaneously, building pattern recognition model based on intelligent algorithm which is suitable for data acquisition can effectively improve the recognition rate and training speed.The main research results of this piece are as follows.?Construction of odor collection platform includes sensor selection, sensor array construction, sampling system and the construction of data acquisition module. The platform has realized odor collection of Staphylococcus aureus, Pseudomonas aeruginosa, Escherichia coli and the mixture of two above compounds based on broth culture medium. In order to improve the degree of difference between samples, some samples were diluted with distilled water, ensuring the accuracy of the extraction of odor characteristics.?Selecting the appropriate pretreatment and feature extraction methods to construct a sample matrix for the training classifier considering to the characteristics of the collected data. Meanwhile, this piece introduces the common algorithm in E-nose, including PCA/KPCA used to select feature, LDA/KDA for feature dimension reduction, support vector machine (SVM) and random forest optimized with QPSO to classify the samples.?Using a combination of various validation methods verifies the classification effect of the classifier on the sample set, including two classification model models for infection and uninfected and multi classification model of different bacterial infection. The experiments show that random forest which is independent of parameter selection is superior to intelligent algorithm based on parameter selection. To a certain extent, the optimization algorithm can make up the deficiency of the intelligent algorithm which depends on the parameter selection to a certain extent. The optimization algorithm combined with a variety of parameter selection algorithm has been unable to exceed random forest in classification performance. As a new classification algorithm, random forest has a broad application prospect and practical value in the E-nose for the detection of wound infection bacteria.