Respiratory Disease Detection using Depthwise Seperable Convolutional Neural Networks

Respiration is the process of converting inhaled oxygen to carbon dioxide by leaving energy as a by-product. Although as scientific as it sounds respiration is the most natural and biological process with a plethora of unpredictable outcomes and diseases. Respiratory diseases originate and spread throughout the pulmonary region of one's body. The diseases are generally very common in patients, but they also depend on climatic and environmental conditions. As of now, respiratory diseases are only being identified after they are spread and affected a vast area of the lungs. And there would be less time for prognosis at the extreme stage. In literature, various methods are proposed for respiratory sound detection and these existing methods are limited in capability of detecting any respiratory disease at an early stage. This paper proposed a new technique to detect the respiratory sound efficiently using a depth-wise separable convolution neural network (DS-CNN). The proposed method accuracy is obtained at 92% and compared with the state of arts.