An Empirical Evaluation of Various Image Segmentation Techniques for Tuberculosis Detection

Abstract

The most important step towards implementing a computer-aided diagnosis system for Chest X-Ray Images is automated tuberculosis segmentation. Tuberculosis is one of the world's most serious public health issues affecting women. Image segmentation techniques aim to identify and extract foreground objects in an image, resulting in individual segments. Image segmentation differs greatly from one type of image to the next because each has its own context and different geometrical properties procedure. An effort is made to compare and study the efficiency of tuberculosis detection segmentation using U-Net, FCN, and Faster R-CNN. The analysis concluded that the performance of segmentation using U-Net yielded better results than other techniques.