Algoritmos de procesado de señal basados en Non-negative Matrix Factorization aplicados a la separación, detección y clasificación de sibilancias en señales de audio respiratorias monocanal

Abstract

Auscultation is the first clinical examination that a physician performs to evaluate the condition of the respiratory system, because it is a non-invasive, low-cost, easy-to-perform and safe method for the patient. However, the diagnosis derived from auscultation remains a subjective diagnosis that is conditioned by the ability, experience and training of each physician in the listening and interpretation of respiratory audio signals. As a result, a high percentage of misdiagnoses are produced that endanger the health of patients and increase the cost associated with health centres. This Thesis proposes new methods based on Non-negative Matrix Factorization applied to separation, detection and classification of wheezing sounds in order to provide a complementary information pathway to the physician that helps to improve the reliability of the diagnosis made by the doctor.