COVID-19 and Respiratory Diseases Classification using Deep Convolution Neuron Network

Abstract

This study proposes COVID-19 and Respiratory Diseases Classification using Deep Convolution Neuron Network. ICBHI 2017 Respiratory Sound Database including COVID-19 from Coswara databased were used in our experiments. The potential results show that the left side model performances are 0.85 accuracy, 0.76 sensitivity, and 0.90 specificity. The right side model performances are 0.86 accuracy, 0.76 sensitivity, and 0.93 specificity. No side set model performances are 0.83 accuracy, 0.71 sensitivity, and 0.93 specificity. In addition, the lung characteristics and lung functions are different among left and right. Therefore, the breathing sound from left and right lung are difference. For this reason, the cross-model performances were evaluated to test this assumption. The cross-model performance results show that the left data is consistent with the left model. As same as the right data is consistent with the right model. Furthermore, the experiment found that mixing training data built the no side set model is the lowest performance. In addition, the proposed framework tends to achieve high performance when compared with a recent study.