Audio based sensing of wheeze in respiratory signals

• Main Author: Seah, Meng Shi.
• Other Authors: Ser Wee
• Format: Final Year Project
• Language: English
• Published: 2011
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation::Medical electronics
• Online Access: http://hdl.handle.net/10356/45868
• Institution: Nanyang Technological University

The purpose of this report is to study the effect on the use of different parameter features performance selected for the detection of wheeze signals. The features selected for this case study are Kurtosis, Renyi Entropy, Mean Crossing Irregularity and f50/f90 ratio. They are selected for their distinct characteristics that allow the evaluation on the behavior of the wheeze signals in both time and frequency domain. This report describes the whole process using the Audio Signal Classification (ASC) block diagram model for illustration. Begin from where the signals are recorded then the filtering of quality signals followed by data extraction using the selected features, and lastly the classification of the classes that permit of dimension reduction from four-dimensional space features to single space so as to allow the projection of an optimal direction w which separate the two classes’ best. The classification is done by the implementation of Fisher Discriminiant Analysis (FDA). The results analyses are made mainly based on the four locations recorded for the wheeze signals, and judging from these locations; at different locations will obtained different detection rates are observed. The rate of success and the false alarm rate are calculated. These values are computed based on the contingency table or confusion matrix to show a better evaluation on the performances which are able to classify the classes accordingly. It had shown that a rate of 100% signals detection at location 3 and 4, and a very low rate of false alarm values are computed at these four locations.