Lung noise recognition employing convolutional neural networks (#642)

Authors

Salamea, Christian

Torres, Mario

Cordero, Mateo

Abstract

Currently, the use of artificial intelligence to aid medical diagnosis is increasingly accepted. This paper presents a new method for identifying the presence of asthma, pneumonia, or COPD in patients with respiratory diseases. The identification is based on auscultation-captured lung noise, using Mel cepstral coefficients and convolutional neural networks for classification. The proposed system achieved optimal performance with a minimum of 15 training epochs and a batch size of 32. It achieved up to 88% accuracy in detecting the aforementioned lung diseases, with a particularly high accuracy of 98.8% in identifying COPD and 54.5% in identifying normal cough.