Construction Of Medical Electronic Nose Prototype And Study On Sensors

Electronic nose,also called artificial olfactory system,has been widely used for the quality control of perfume, beverage, food, etc.. However, the medical electronic nose used for disease diagnosis stays immature,still under studying in laboratory. This project studied and designed a prototype of medical electronic nose,the key points of research work were gas sensor selection, sensor array design, temperature drift counteraction in sensor array and sample input system.This thesis demonstrated the feasibility of medical electronic nose from the pathology point of view,and finally aimed at the examination of lung cancer. Current available types of gas sensors were studied, such as MOS and QCM, etc.. Finally five MOS type gas sensors, TGS822,MQKII,TGS2620,TGS2602 and QS-01 were selected to form up sensor array in our system to detect the volatile biomarker in patients' exhalation.As the concentration of volatile biomarkers in patients' exhalation is very low (in ppb to ppt level), very low level of LoD is required for sensors used in electronic nose while current common gas sensors can not reach such a LoD. This thesis did not only set forth the indispensability of pre-concentration, but also build an experiment scheme to counteract the temperature drift with an eye to improve the performance of sensor array itself. The basic idea of the scheme was to add temperature sensor in the chamber, whose output would be utilized to help sample recognition, and also sample input method of headspace with carrier gas was chosen, some other measures were also taken to improve gas tightness and airflow uniformity of the chamber.In our experiment, the 0.1% and 0.5% ethanol solution were prepared and purged by the carrier gas with the same flow, then two kinds of gas samples were worked out and pumped into the chamber, where the sensor array lies. Twenty samples were eventually obtained by the experiment (ten for each concentration).During the data analysis, the output signals of some sensors were not stable enough as the gas flow was not uniform, the multi-point moving average technique was employed to pretreat the original signal from sensors. Some features were extracted over the smoothed signals. Finally, the maximum value, the relative value and the response time of sensors were individually taken as features. Meanwhile, the perceptron and Back-Propagation Neural Network (BPNN) were built. The former 16 of 20...