Ensemble deep learning method for Covid-19 detection via chest X-rays

Coronavirus disease 2019(COVID-19), a contagious disease that appeared at the end of 2019, has become a significant problem worldwide. It has already claimed the lives of millions of people around the world. Early diagnosis of COVID-19 can significantly decrease the transmission rate and prevent the virus from spreading around. In this paper, we present deep learning-based methods to detect COVID-19 from given chest X-rays( CXR) image of patient. First, we trained three states of the art algorithms(two efficient net with different initial pre-trained weight and SE-ResNext) that can classify a given X-rays images into three classes: covid, pneumonia, and normal separately. Finally, we ensembled these three algorithms by averaging their outputs for each class to obtain a robust prediction. We evaluated our proposed ensembled method on the Postgraduate (PG) Challenge - COVID-19 Detection via Chest X-rays database with 17958 CXR images for training, 3432 for validation, and 1200 for testing. The experimental results show that the proposed method achieves promising performance with an accuracy of 0.9592, a sensitivity of 0.9592, specificity of 0.9597 on the given Test Set Leaderboard. These promising results demonstrate the potential of deep learning-based methods and the importance of ensemble and transfer learning for given CXR images.