Auto Thresholding Sputum Color Image Segmentation For Tuberculosis Diagnosis Base On Intuitionistic Fuzzy

Abstract

In this paper, an automatic algorithm for detecting the number of Mycobacterium tuberculosis is presented from the AFB smear image on I and V-shaped colonies, by applying fuzzy Intuitionistic based on the auto-thresholding segmentation method. Acid-fast bacteria, hereinafter referred to as AFB, are a group of bacteria that have unique characteristics, namely that they can prevent acid decolorization during the staining process, so that when sputum preparations are given a blue color, the AFB will retain its red color. One of the main problems in detecting the number of bacteria based on AFB segmentation is due to differences in light intensity and contrast (due to different lighting distributions). This study aims to segment the AFB images data as a whole, without dividing 1 bacterium into several parts. The segmentation process uses the stages of patch preparation, mask preparation and UNet Architecture. In mask preparation process, it is compare 3 color threshold models (grayscale then black and white - Adaptive Histogram then black and white - Fuzzy Intuitionistic- then black and white), all three are then segmented using the UNet method. The novelty of this paper is the creation of an input image mask. In this research, an optimization method is used with a maximal entropy approach. The idea is to find the maximum degree of disorder by calculating the entropy on the modified input image matrix. From the experimental results, it was found that the method that has high accuracy in segmenting AFB on I and V-shaped colonie is the fuzzy intuitionistic method, with an accuracy rate of 94.78%.