COVID-19 detection for chest X-ray images using local binary pattern

Abstract

The increase in patients with COVID-19 is overwhelming in healthcare systems around the world. Due to the large number of people affected by this pandemic, the medical and healthcare departments are facing a delay in the detection of COVID-19. Besides, it is not an easy task to clarify the images from the radiograph on what types of infection between bacteria pneumonia and COVID-19. The automatic feature analysis can help physicians more precisely in the treatment and diagnosis of diseases. In this research, Local Binary Pattern (LBP) texture features algorithm has been proposed to automate the current manual approach. This process starts by extracting the intensity grayscale texture from the normal, bacteria pneumonia and COVID-19 chest x-ray images. To prove the accuracy of LBP, a simple classifier k-Nearest Neighbour (k-NN) has been implement to classify the chest x-ray images into normal, bacterial and pneumonia class. The 10-fold cross validation has been used to validate the chest x-ray images. From the evaluation, 96% accuracy can be achieved by using LBP as a feature extraction feature. It shows that LBP is a powerful texture features to detect COVID-19 from the x-ray images. More samples will be collected in the future and neural network approach is suggested as a classifier in the future due to its ability to imitate human respond.