Automatic assessment of the degree of TB-infection using images of ZN-stained sputum smear: New results

We present new results in the context of automatic assessment of the presence of acid fast bacilli (AFB) in images of ZN-stained sputum smears. Specifically, the first phase involving color segmentation in the HSV space is improved in terms of quality by using a decision-Tree classifier. Further, we have recognized the possibility of staining artifacts of large size, and propose a method of discriminating the same from clumps of AFB. The method involves the use of Haralick's texture features. Its importance lies in the fact that the presence of large clumps or even several small clumps in an image of a sputum smear generally indicates a higher degree of infection. The results of segmentation-as assessed by the Sorenson-Dice coefficient & the Hausdorff distance-are better than those pertaining to our previous work. The counts of AFB are close to those based on visual inspection, and the clumps could be separated from large staining artifacts successfully.