Neural network classification software for the gas analytical survey data of respiratory system

An algorithm and neural network architecture are proposed for classifying signal patterns generated in devices for analyzing the composition of a gas mixture in exhaled air. The devices use a set of non-selective semi-conductor gas sensors synchronized with each other and operating in the thermal cycling mode. The implemented algorithm and neural network provide for the normalization of signal values simultaneously from the entire set of sensors and the differentiation of signal patterns. The software package is implemented in the C++ algorithmic programming language in the Qt environment and allows for additional training of the neural network by integrating tools from the database management system and neural network data analysis when increasing the database volume. The studies used data from exhaled air samples taken from healthy volunteers and patients with squamous cell carci-noma of the larynx and oropharynx. The obtained sensitivity and specificity indicators of the neural network data classifier are comparable with modern high-precision X-ray methods for diagnosing respiratory tract tumors.