Classificador de sons pulmonares: uma abordagem baseada em FFT e maquina de vetor de suporte

Abstract

The diagnosis based on pulmonary auscultation is a routine activity in medical care. However, it is a highly dependent user technique and requires a quiet ambient. In this sense it is interesting that there are systems capable of assisting the medical diagnosis in relation to the auscultation sound. In this way several works using computational intelligence have been made for the processing of these pulmonary sounds and thus, together with the user experience, make possible a more reliable diagnosis. This work presents an approach to classify pulmonary sounds initially between vesicular and adventitious, and adventitious sounds are classified into six classes: monophonic and polyphonic wheeze, coarse and fine crackles, stridor and pleural rubs. The approach is based on Fast Fourier Transform (FFT) and Vector Support Machine (SVM). To extract the characteristics of the sounds, we used the FFTs which are then evaluated by means of a Genetic Algorithm that works in accordance with the best linear classification kernel SVM. It was possible to classify pulmonary sounds among the seven classes with results from 93, 3 ± 1, 6% to 100, 0 ± 0, 0%.