Research on Classification of Respiratory Diseases Based on Multi-features Fusion Cascade Neural Network

Abstract

Respiratory diseases have a significant impact on the health and social economy of the population, and there are currently limited ways to detect respiratory diseases in hospitals. To this end, we proposed a cascade neural network model based on multi-features fusion to classify respiratory diseases. Meanwhile, we also used two different pre-processings to input respiratory sounds into three different deep neural networks for comparative experiments. In order to solve the problem of class- imbalance of the dataset, we extend the dataset. Our system classifies six respiratory diseases, and achieves 88.3% ICBHI average accuracy, respectively. The average accuracy is repeated on ten random splittings of 80% training and 20% testing data using the ICBHI 2017 dataset of respiratory cycles.