Sound based wheeze signal detection using ELM algorithm

• Main Author: Udomsangpetch, Karn
• Other Authors: Huang Guangbin
• Format: Final Year Project
• Language: English
• Published: 2012
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: http://hdl.handle.net/10356/49708
• Institution: Nanyang Technological University

Wheezing is a prevalent symptom found in most patients suffering from obstructive respiratory disease such as asthma. Traditional methods used in identifying wheezes are mostly manual and heavily rely on the physician subjective judgment. Not only do such methods prove to be time-consuming and inefficient, but they are also inaccurate at times due to human errors and subjectivity. Various automated methods were introduced for the past few decades but due to irregularities in the occurrence and nature of wheezes, many difficulties were encountered. In order to find complex relationships among the data and create an adaptable prediction model, Extreme Learning Machine (ELM), a type of learning algorithm for artificial neural network, is applied to the breathing samples obtained from real patient. By forming a suitable training data set for ELM to learn from after a careful analysis and observation of the waveforms in time and frequency domains, the testing samples can be classified into wheezes signal and normal breathing sound with top sensitivity of 72.60% and specificity of 84.80%. Another method using support vector machine (SVM) also yields roughly the same result of 70% in sensitivity and 80% in specificity. Further improvement needs to be done on the formation of the training data to increase the accuracy of the machine classification.