Research On The Key Technology Of Medical Electronic Nose

The medical electronic nose is a special e-nose system for disease diagnosis. It is mainly composed of a gas sensor array, a signal preprocessing unit and a pattern recognition algorithm. With the characteristics of being non-invasive, convenient and efficient in disease diagnosis, e-nose has become an attractive diagnostic method.Applying the medical e-nose in breast cancer diagnosis and wound pathogen detection, this thesis investigates three aspects of the medical e-nose: sensitivity, repeatability and differentiating capability. The main work includes:1. Because the detectable concentration range of the current gas sensors can not satisfy the requirements of clinical test for the medical e-nose, we have designed a solid trap/thermal desorption-based gas condensation system which consists of a flow control unit, a temperature control unit and a sorbents-filled trap tube. The gas condensation system together with the medical e-nose improved the system's sensitivity and anti-interference capability, taking a preliminary step to solve the problem that the detection capability of the current gas sensors is not strong enough to enable the e-nose to be directly used in diagnosis of exhaled breath disease.2. The effectivness and repeatability of experiments on e-nose are closely related with the system parameters. In traditional e-nose experiments, parameters are set by experience, failing to guarantee the repeatability and maximum effectiveness of the experiment. Our work has introduced the design of experimental method into the research and development of e-nose, and systematically studied the influence of the operation parameters of e-nose and pre-condensation system on the experimental results. First, the total variation of baseline was taken as the indicator and the fractional factorial method was adopted to study the effect of each controllable factor on the experimental results, and the main effect and interaction effects of various controllable factors on the total baseline variation were analyzed, so that we can consider the main effect in system construction, optimizie the system configuration. Then, the response surface methodology was used in the analysis of the influence of the key factors on the gas condensation system to achieve the best condensation effect and guide the parameter setting in the condensation system.3. Repeatability is a key issue in the research of medical e-nose. This work gives the definition of repeatability of the medical e-nose and examines the main factors that influence the repeatability of the medical e-nose: experimental errors, noises, sensor drift and environment. And also, this work presents the qualitative and quantitative indicators to evaluate the repeatability of the medical e-nose: box plots and repeatability score, with which the repeatability of the sensor array, signal preprocessing method and feature extraction method were studied.4. The optimization of the medical e-nose array is to eliminate redundancy and correlation and enhance the differentiating capability of the system. Based on the analysis of the common optimization method of gas sensor array, we have proposed a sensor-array optimization method based on adaptive genetic algorithm for the medical e-nose, which realizes the elimination of redundancy and correlation and achieves the optimization of sensor array by setting the significant coefficients of the sensors. The new array optimization method was used to detect five breast cancer character gases and seven wound pathogens. Results show that the new method can effectively optimize the array and enhance the differentiating capability of medical e-nose system. The traditional array optimization method can be regarded as a special case of the new one with respect to the significant coefficient. The traditional method without or with the corresponding sensor corresponds to the new method with the significant coefficient set to be 0 (removing the corresponding sensor) or 1 (retaining the corresponding sensor) respectively.5. An appropriate feature extraction and selection method is crucial to enhance the differentiating capability of the medical e-nose. Based on the traditional feature extraction method, this work has proposed a feature extraction method based on wavelet transform and a feature selection method based on the scatter matrix. One dimensional discrete wavelet transform of the original response curves of the gas sensors was carried out and some wavelet coefficients were selected preferentially as the feature of classifier according to the scatter matrix. Experiments on the detection and identification of five breast cancer character gases and seven wound pathogens show that the new feature extraction and selection methods can increase the differentiating capability of medical e-nose.6. Further, this work analyzed the fundamental principle of the pattern recognition algorithm based on empirical risk minimization, and confirmed that the probabilistic neural network classifier, under the premise that the a priori probabilities of all classes are equal, is equivalent to the Bayes classifier based on the kernel function, and can achieve optimal classification with the minimum errors. The identification results of the single component, mixed component and drift-interferenced component of the wound pathogens show that the probabilistic neural network classifier can achieve the accurate classification of the wound pathogens and has a strong ability to restrain the drift. Because the pattern recognition of the medical e-nose is an asymmetric system with small size samples in nature, the traditional classifier based on empirical risk minimization may not work satisfactorily. This work discusses the possibility of the application of support vector machine (SVM) based on statistical learning theory in the pattern recognition of medical e-nose.