Detection of mycobacterium tuberculosis in tissue using k-Nearest neighbour and fuzzy k-Nearest neighbour classifiers

Abstract

Early detection and treatment are the most promising way to increase a patient's chance of survival from TB disease, reduce the duration and cost of treatment, and prevent the disease from spreading. Currently, microscopic examination of clinical specimens by medical technologists is the most widely used for TB screening and diagnosis. Unfortunately, the process is tedious, timeconsuming and error-prone. This paper describes a method for automated TB detection from tissue sections using image processing techniques and artificial intelligence. The proposed work consists of three stages; image segmentation, features extraction and classification. Tissue slide images are acquired using a digital camera attached to a light microscope. Then, k-mean clustering and thresholding techniques are applied for image segmentation. The segmented regions are further classified into three classes; ‘TB’, ‘overlapped TB’ and ‘non-TB’. A set of six geometrical features; area, perimeter, shape factor, minimum and maximum distance of a pixel in the boundary from the centroid, and eccentricity, are calculated from the segmented regions to describe their shape properties. Finally, k-nearest neighbour (kNN) and fuzzy k-nearest neighbour (fuzzy kNN) classifiers are used to classify the feature vectors. The experimental results suggested that the kNN classifier performed slightly better than the fuzzy KNN in classifying the TB bacilli.