Detection of tubercle bacilli in Ziehl-Neelsen stained sputum slide images using Hu’s moment and HMLP network

Abstract

Manual screening by light microscopy is the most widely used method for tubercle bacilli detection in the developing country. However, it is a time consuming and labour-intensive process. In this paper, a method using image processing technique and neural network classification has been proposed for automated tubercle bacilli detection in sputum slide images. The method mainly consists of three main stages: image segmentation, feature extraction and identification. Hybrid multilayered perceptron (HMLP) network using modified recursive prediction error training algorithm have been used to perform TB identification. Experimental results demonstrated that the HMLP network achieved the classification accuracy with percentage of 73.9%.