Tuberculosis Detection using Gray Level Co-Occurrence Matrix (GLCM) and K-Nearest Neighbor (K-NN) Algorithms

Abstract

Research has been conducted on detecting tuberculosis (TB) using machine learning. In this study, chest Xray (CXR) image data was used with a pixel value of 512 x 512 and PNG format consisting of normal lung images and TBinfected lung images in a 50:50 ratio; the number of images was 200 training data images and 80 testing data images. In the preprocessing stage, grayscaling is carried out so the image has a grayscale. Then, do the image improvement using contrast stretching. Furthermore, image extraction was carried out using 22 GLCM features with variations in the direction of the angles of 0°, 45°, 90°, and 135°. The result of feature extraction data is then identified using KNN Classification. The training results have the highest accuracy value with variations in the direction of the GLCM angle of 45° and the value of K = 3; at the testing stage, it produces an accuracy of 90%.