A new pulmonary disease diagnosis system based on EfficientNet and transfer learning: pulmonary disease diagnosis based on EfficientNet and TL

Abstract

Pulmonary epidemic diseases are one of the main causes of human death. Pulmonary epidemic diseases are usually highly contagious because they can be transmitted by droplets. In this study, we mainly focus on two types of common pulmonary epidemic diseases: COVID-19 and tuberculosis. COVID-19 has spread all around the globe since December 2019. The widespread COVID-19 caused the lockdown of the cities and economic losses. On the other hand, tuberculosis is among the ten highest human killers. Accurate and rapid diagnosis of pulmonary epidemic diseases is the primary step in clinical treatment. Therefore, we propose to leverage deep learning models to identify pulmonary epidemic diseases based on chest computed tomography (CT) images. We select the EfficientNet as the backbone model and employ a transfer learning method to train the model on our chest CT dataset. Experimental results reveal that our method can achieve promising classification performance, which is comparable to state-of-the-art approaches.