Design of a system for diagnosis of lung diseases using pulmonary sounds

Auscultation is a common method among physicians to detect adventitious sounds which are mainly wheezes and crackles, and are important indicators of a pathological condition. According to the presence of these adventitious sounds doctors diagnose a patient as healthy or pathological and may require further examinations. However, auscultation is highly subjective method and there is always a risk of missing adventitious sounds or misevaluating them. In this thesis, a new system is designed and implemented to reduce this risk by removing subjectiveness of the method using computerized techniques. Another advantage of this system is the elimination of unnecessary examinations and the saving of time for both doctors and patients. This system is intended to be used for detection of pathological conditions in recorded sounds rather than exact diagnosis of respiratory diseases. This system is designed to determine the presence of any respiratory disorder in a subject. For this purpose, the system analyzes recorded pulmonary sounds with digital signal processing tools and machine learning techniques. It consists of three different detection algorithms and a final classifier to combine outputs of all algorithms to produce a result. These algorithms are wheeze detection algorithm, crackle detection algorithm, and disease detection algorithm.