A computer assisted diagnosis system for the identification/auscultation of pulmonary pathologies

• Main Authors: Cordel, Macario O., II, Ilao, Joel P.
• Format: text
• Published: Animo Repository 2016
• Subjects: Lungs—Sounds; Auscultation; Lungs—Diseases—Diagnosis; Pattern recognition systems; Support vector machines; Computer Sciences; Respiratory Tract Diseases
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/5438; https://www.researchgate.net/publication/283086929_A_Computer_Assisted_Diagnosis_System_for_the_IdentificationAuscultation_of_Pulmonary_Pathologies
• Institution: De La Salle University

Statistics show that the primary cause of morbidity and mortality among Filipinos are pulmonary illnesses. These illnesses could have been prevented if detected and treated early. With the physicians medical knowledge and experience, early detection of possible common pulmonary diseases can be performed using a stethoscope. However, with the current physician-to-population ratio in the country, early detection of respiratory diseases may not be performed on most cases especially in the rural areas, causing even benign cases to lead to mortality. In this paper, we present the development of a system that classifies lung sound for possible pulmonary pathology.Using an electronic stethoscope, lung sounds were collected from healthy individuals and patients with common pulmonary problems for the developed systems training and evaluation. The collected data were pre-processed in order to remove mechanical and other external noises. Using Support Vector Machine (SVM) for modelling and classification, the developed system was able to achieve 100% identification of the normal lung sound from the adventitious lung sound, with an average cross-validation performance of 88%. The developed system, however, has low performance in classifying specific lung sounds, that is, normal vs. crackle vs. wheeze vs. ronchi, with an average accuracy of 61.42% and an average cross-validation performance of 90%.