Detecting and Counting Tuberculosis Bacilli in a Microscopic Image of Patient Sputum Using Gaussian Mixture Model

Diagnosis of Tuberculosis (TB) based on a direct microscopy observation is a challenging task due to, among others, varying qualities of the sputum preparations. Even to determine a TB-negative patient, WHO stipulates that 100-300 non-overlapping fields of view must be carried out. This is undoubtedly tiring work to do, so that it can lead to misdiagnosis in patients. This paper proposes a method to assist physicians in analyzing TB sputum preparations based on digital microscopic images of ZN-stained preparations (Ziehl-Nielsen). This study proposes using the Gaussian Mixture Model (GMM) algorithm and Markov Random Field (MRF) parameters to segment the TB result image from its background image and count the number of TB yields in the image. This method was chosen because it has a fast computation time and can handle a limited number of image datasets. The data used in the form of 32 training data and 8 test data taken randomly from the Ziehl- Neelsen Sputum Smear Microscopy Image Database (ZNSM- iDB) dataset. From the results of tests carried out on 8 data, the system shows that the level of accuracy, precision, sensitivity, and specificity of the system are 93.52%, 97.53%, 91.87%, and 83.16%, respectively.