COVID-19 Detection in CXR Image Using High Frequency Emphasis Filtering Based Convolutional Neural Network

As a huge disaster for humanity, the COVID-19 has caused many negative effects on the lives of people around the world with a rapid growth. Moreover, the global pandemic of Neocoronavirushas produced many mutated strains. Although the most commonly used test for COVID-19 is reverse transcription-polymerase chain reaction (RT-PCR), CXR becomes an irreplaceable tool for the diagnosis and analysis for a more complete and accurate visualization of the lung lesion process. Therefore, it is of high value for classification and identification studies. In this paper, the high-frequency emphasis filtering based convolutional neural networks (HFEF-CNN) are proposed for solving the automatic detection of COVID-19. Firstly, the HFEF is used to denoise the image data to make some features in the image more obvious. Then some major CNNs are used to train image classification models to achieve better detection performance. Finally, Some experiments are conducted on the “COVID-19 Chest X-Ray Database” dataset. To verify the effectiveness of the HFEF-CNN, a histogram equalization based CNN (HE-CNN) and a restricted contrast adaptive histogram equalization based CNN (CLAHE-CNN) are compared. The experimental results show that the HFEF-CNN outperformed the above two methods.