Multiclass Categorisation of Respiratory Sound Signals Using Neural Network
Naseem Babu, Jyoti Kumari, Jimson Mathew, Udit Satija, Arijit Mondal
Abstract
Respiratory diseases have seriously impacted human life in the last couple of years; as Covid 19 arrived, many lost their beloved ones. Since respiratory diseases directly attack the patient’s lungs, it is becoming risky day by day for human life and doctors because a confined number of resources are available in hospitals to detect these respiratory diseases, and detection of these diseases is a difficult job to the doctors. Therefore early-stage diagnosis can help the doctor in saving human lives. Researchers are continuously trying to help doctors by designing efficient and more accurate tools for detecting different types of respiratory diseases. This paper uses a convolution-based deep learning model to classify these respiratory diseases using patient respiratory sound signals with Mel frequency cepstral coefficients (MFFCs) as a feature vector. In this paper, we have tried to keep our neural network model as simple as possible with less trainable parameters and good classification accuracy. The model performance is measured in terms of sensitivity, specificity, average score, and harmonic score.