Computerized diagnosis of respiratory disorders. SVM based classification of VAR model parameters of respiratory sounds.

Abstract

This article is part of the Focus Theme of Methods of Information in Medicine on "Biosignal Interpretation: Advanced METHODS for Studying Cardiovascular and Respiratory Systems". This work proposes an algorithm for diagnostic classification of multi-channel respiratory sounds. 14-channel respiratory sounds are modeled assuming a 250-point second order vector autoregressive (VAR) process, and the estimated model parameters are used to feed a support vector machine (SVM) classifier. Both a three-class classifier (healthy, bronchiectasis and interstitial pulmonary disease) and a binary classifier (healthy versus pathological) are considered. In the binary scheme, the sensitivity and specificity for both classes are 85% ± 8.2%. In the three-class classification scheme, the healthy recall (95% ± 5%) and the interstitial pulmonary disease recall and precision (100% ± 0% both) are rather high. However, bronchiectasis recall is very low (30% ± 15.3%), resulting in poor healthy and bronchiectasis precision rates (76% ± 8.7% and 75% ± 25%, respectively). The main reason behind these poor rates is that the bronchiectasis is confused with the healthy case. The proposed method is promising, nevertheless, it should be improved such that other mathematical models, additional features, and/or other classifiers are to be experimented in future studies.