A topological approach for the pattern analysis on chest x-ray images of COVID-19 patients

Due to the global outbreak of COVID-19, in this work, anterior-posterior (AP) and posterior-anterior (PA) chest X-Ray images were used as the input data for computational image processing. This to approximate a range of luminescence that could filter the anatomical region of the lungs, by comparing local maxima in the luminescence histograms obtained from an open dataset of chest X-Ray images stored in a public GitHub repository at https://github.com/ieee8023/covid-chestxray-dataset. Luminescence masks were obtained from the approximated values of luminescence in the image histograms that correspond to the anatomical region of the lungs in the original chest AP and PA X-Ray images. The luminescence masks were used to segment the regions of interest containing the lungs, storing them in a separate image. The luminescence histograms from the segmented images were given as inputs for the K-means algorithm; a non-supervised learning algorithm that was applied as part of the pipeline of the mapper algorithm to obtain groups of information in data in the process of clusterization. The mapper algorithm provides a visual representation of the patterns found in clusters obtained from the values of luminescence frequency in the images through interconnected nodes in a simplicial complex. A simplicial complex is a mathematical object that allows observing topological features in a graph created by nodes connected by edges. Mapper algorithm closely connects regions of nodes in the simplicial complex, it indicates ranges of luminescence values in the input images which provide helpful information in the analysis of chest X-Ray images