Pneumonia Detection Electronic Nose Device And Algorithm Research

Intensive Care Unit(ICU)is an important part of medical treatment in the hospital.However,the patients are easier to be infected by the pneumonia after using the mechanical ventilation breathing machines,which called ventilator-associated pneumonia(VAP).Concerning that patients are not real-time detection of pneumonia infection status after using the ventilator in ICU,a new solution for VAP diagnose was proposed.In this thesis,Cyranose-320 electronic nose was used to collect the signal from exhaled gas by the patients.Artificial neural network(ANN)and support vector machine(SVM)were used to build bacteria recognition models.Cross-validation was used to evaluate the stability of both models.Final,the ensemble method was used to rebuild the ANN model.The results show that both ANN and SVM models have high recognition of pneumonia.The accuracy(ACC)are 0.9141±0.0313,0.8753±0.0389 respectively,the sensitivity(SEN)are0.9292±0.0553,0.8839±0.0585 respectively,and the positive predictive value(PPV)are0.8960±0.0242,0.8693±0.0240 respectively.The test results show ANN model have a little better performance than SVM model in recognition of pneumonia.Finally,the ensemble neural network(ENN)model is built.The results show that comparing with the traditional one,the ANN model which built by ensemble method have a little better performance in ACC is 0.9277±0.0170,SEN is 0.9512±0.0451,PPV is 0.9110±0.0355 respectively.This study aims to predict patients whether infected with pneumonia or not through ANN,SVM and ENN models,providing a scientific and effective reference for physician to perform early diagnosis.