Image processing techniques for identifying Mycobacterium tuberculosis in Ziehl-Neelsen stains.

Worldwide, laboratory technicians tediously read sputum smears for tuberculosis (TB) diagnosis. We demonstrate proof of principle of an innovative computational algorithm that successfully recognizes Ziehl-Neelsen (ZN) stained acid-fast bacilli (AFB) in digital images. Automated, multi-stage, color-based Bayesian segmentation identified possible 'TB objects', removed artifacts by shape comparison and color-labeled objects as 'definite', 'possible' or 'non-TB', bypassing photomicrographic calibration. Superimposed AFB clusters, extreme stain variation and low depth of field were challenges. Our novel method facilitates electronic diagnosis of TB, permitting wider application in developing countries where fluorescent microscopy is currently inaccessible and unaffordable. We plan refinement and validation in the future.