Investigation of an appropriate feature extraction methods based on single channel lung sounds

Abstract

In the past, multichannel recording methods often used in the studies for automatic recognition of lung sounds did not fit the lung auscultation. Due to these studies performed on different standards, automatic diagnostic methods for lung sounds have not been developed until now. For this reason, more work is needed to develop a suitable automatic recognition method for single-channel lung sounds. Nowadays, thanks to the advanced electronic stethoscope that is suitable for the auscultation procedure, lung sounds can be recorded as single channel. In this study, an effective feature method was investigated in order to classify commonly heard and single channel recorded lung sounds with high accuracy. In the classification phase, the results are examined in our work of many of the features extract from the time and frequency domain and the Mel Frequency Cepstrum Coefficients using Naive Bayes, Linear Discrimination Analysis and Support Vector Machines. As a result, the most efficient feature was obtained when using features extracted from frequency domain and Mel Frequency Cepstrum Coefficients.