Automatic identification of mycobacterium tuberculosis from ZN-stained sputum smear: Algorithm and system design

Tuberculosis (TB) is a communicable disease for which early diagnosis is critical for disease control. Manual screening for TB identification involves a labor-intensive task with poor sensitivity and specificity. To improve the diagnostic process we develop an automated system for TB identification, which consists of an automatic microscope, an image-based autofocus algorithm and an image-based TB identification algorithm. The system can automatically capture a large number of clear images on sputum sample and process all the images in real time to identify the bacilli and count their number. In order to speed up image acquisition while guaranteeing the image quality, an efficient method for capturing the images is proposed. To obtain fine segmentation results, a two-stage segmentation method based on both the HSV and CIE L*a*b* color space is developed. To identify the TB bacilli, the algorithm uses three shape feature descriptors, which are area, compactness and roughness, and makes the judgment using a decision tree. Experimental results confirmed the superior performance of the proposed algorithm.