Polyaniline nanocomposites based sensor array for breath ammonia analysis. Portable e-nose approach to non-invasive diagnosis of chronic kidney disease

•An electronic nose has been developed based on polyaniline nanocomposite sensor array.•Concentration ranges of ammonia can be discriminated in a simulated human breath.•Using a sensor array reduces drift, repeatability or fault hazards associated with the use of a single polyaniline sensor.•Selecting the right features and classification algorithm improve the diagnosis accuracy.•Such a system could allow a non invasive diagnosis tool of chronic kidney diseases. Kidney failure is a serious chronic disease, defined as the irreversible loss of kidney function. This disease is clinically silent to a very advanced stage. Thus, only a screening procedure can diagnose its pathology early enough to slow its progression. Due to the known fact that pathology of this disease is characterized by an increase of ammonia concentration in breath, its monitoring with a portable system can be a simple way for a noninvasive and early diagnostic on site. To realize such a system a new specific conductometric array of 11 different polyaniline nanocomposite sensors is used, based on the electronic nose principles. This approach allows bypass sensor weaknesses (sensor drift and sensitivity to humidity) and to determine ammonia in the typical concentration range of human breath (500 ppb–2100 ppb). In particular, polyaniline based nanocomposites with either titanium dioxide, chitosan or carbon nanotubes are used to provide different sensitivities and response times. This allows associating a single pattern of sensor responses to a concentration range. Maximum variation of resistance, derivative and integral values are extracted from the response curves of each sensor. Common classifiers are then tested and a selection feature algorithm is used. It permits improving the measurement accuracy and determining the most relevant features and sensors. Diagnosis accuracy reaches 91% with the combination of feature selection and Support Vector Machine algorithm.