A new algorithm for automatic assessment of the degree of TB-infection using images of ZN-stained sputum smear

This paper describes a new algorithm for automatic assessment of the degree of TB-infection, by counting the number of Mycobacteria i.e., acid fast bacilli (AFB) in the color images of ZN-stained sputum smear. This algorithm consists of two stages. The first (“pre-processing”) stage involves exploiting color information to segment the candidate AFB in the image from the background, based on classification of pixels in the HSI color-space using Mahalanobis distance. In this context, we introduce a novel “divide and conquer” strategy to improve the robustness of color-classification. The pre-processing stage is followed by connected component labeling, size-thresholding to remove noisy objects, proximity-grouping by using a novel proximity-test algorithm, and area-based classification. Our algorithm identifies and counts the number of AFB irrespective of their shapes, can handle bacilli with beaded structure (which are important and are specific to TB) and has shown reasonable success in isolating clumps. A total of 205 images of ZN-stained sputum smears taken from more than 12 patients were considered in our experiments. Results on 169 images, based on HSI clusters built from 36 other images, are encouraging. Some of the images used in building the data-base and also in validation, include those sent by the RNTCP (Govt. of India) for training-purposes.