Entrenamiento de una red neuronal convolucional por transferencia de conocimiento para la clasificación de sonidos respiratorios

Abstract

Health is one of the most relevant elements for the development of quality life, being healthy should be part of the overall lifestyle and that is why one of the Sustainable Development Goals (SDGs) is to ensure healthy lives and promote wellbeing for all at all ages. The diagnosis of diseases is an inferential process, it starts with the anamnesis to obtain a clinical picture, aimed at defining the possible disease, this makes it a complex task, in respiratory diseases have as tool auscultation that depends largely on the experience and acoustic training that is limited to a subjective judgment of physicians or examinations with sophisticated and expensive tools such as chest X-ray. This paper presents an approach to perform breath sound classification using Deep Learning models that extract important features from spectrograms of an audio signal that could help physicians recognize abnormal lung sounds making diagnosis more affordable and accessible.