Detection of tubercle bacilli in Ziehl-Neelsen stained tissue slide images using Hu’s moment invariants and artificial neural network

Abstract

Early detection of tuberculosis infection is the key to successful treatment and control of the disease. Manual screening by light microscopy is the most widely used for tubercle bacilli detection but it is time consuming and labour-intensive process. This paper describes a method using image processing and neural network for automated tubercle bacilli detection in tissues. The proposed work consists of three main stages: image segmentation, feature extraction and identification. First, images of Ziehl-Neelsen stained tissue slides are acquired using a digital camera attached to a light microscope. To isolate tubercle bacilli from its background, moving k-mean clustering that uses C-Y colour information is used. Then, seven Hu’s moment invariants are extracted as features to represent the bacilli. Finally, based on the input features, multilayer perceptron network is used to classify into two classes: ‘true TB’ and ‘possible TB’. Six types of training algorithms are used to evaluate the network performance. Experimental results demonstrated that the MLP network trained by Levenberg-Marquardt has achieved the highest accuracy with percentage of 88.57%.