A fusion of fuzzy logic induced local binary pattern for COVID-19 detection from chest X-ray images: interpretation using layer-wise relevance propagation

According to current medical research, COVID-19 induced patient mortality may be reduced from early identification through precise diagnosis, especially in asymptotic cases. In the proposed work, a machine learning model is designed using fuzzy logic and local binary pattern of chest X-ray images to generate the fuzzy local binary pattern (FLBP) to extract the features of the X-ray images. The proposed convolutional neural network (CNN) model is trained employing these FLBP features in order to determine whether the subject having that X-ray image is healthy or COVID-19-affected. The model is trained with three datasets: Dutta_Saha_Saha, COVID-19 radiography database and novel COVID-19 chest x-ray repository achieving accuracy of 97.17%, 96.82% and 97.20% respectively. Additionally, using layer-wise relevance propagation, a relevance map of the FLBP image is created, emphasizing the key features of the image that are utilised to predict the class of the chest X-ray images using the proposed CNN classifier.