Yusuf A. Amrulloh

HMM Based Cough Sound Analysis for Classifying Pneumonia and Asthma in Pediatric Population

Separating pediatric asthma from pediatric pneumonia is one of the major issues in remote areas. These diseases have overlapping symptoms, but require drastically different treatments. Existing guidelines for pneumonia classification in resource poor regions from The World Health Organization call for the use of bronchodilator test to separate asthma from pneumonia. However, bronchodilator is an expensive test to conduct and not easily available in remote areas. In this study, we propose an innovative and novel technique using cough sound analysis to separate pneumonia cases from asthma. In the work of this paper we analyzed cough sound data from 20 subjects (10 pneumonia and 10 asthma patients). Using mathematical features of cough sounds, an HMM classifier was trained to identify pneumonic cough and asthmatic cough. Then by computing Pneumonic Cough Index each patient was classified as either into pneumonia or asthma. Proposed method achieved an accuracy of 90% (sensitivity = 100% and specificity = 80%) in classifying pneumonia and asthma patients. Our results indicate that cough sound carry critical information which can be used to separate asthma patients from pneumonia. Proposed technique in this paper shows potential to become an alternative for bronchodilator test in the resource poor areas of the world.

Hidden Markov model of cough from pediatric patients with respiratory infections

Cough is one of the early symptoms of the respiratory tract infections. Cough sound may indicate the physiology of respiratory tract impairment due to the infections. Inflammation, obstruction and excessive mucus may generate specific types of cough sound. In pediatric population, their cough sound may relate to the etiology of the respiratory diseases. Therefore, cough sound is very useful to support the diagnosis. In the physical examination, physicians may assess cough by listening to several episode of cough sounds. This process is similar to the way human recognize speeches. In this paper we present our work on the development of cough model using a Hidden Markov Model (HMM). The data for this work were collected from pediatric population diagnosed as pneumonia and asthma. Our developed model achieved the accuracy of 82.7% and 52.6% for pneumonia and asthma, respectively. It shows that HMM can be used to model different types of cough from respiratory diseases.