Image processing and neural computing used in the diagnosis of tuberculosis

A method far automating the detection of tubercle bacilli in sputum specimens is described. A fluorescence microscope with an attached digital camera is used to manually locate and capture images of tubercle bacilli. The method comprises two phases: (a) image processing and analysis techniques are applied to the images for enhancement and feature extraction; (b) object recognition techniques are used for the automatic identification of tubercle bacilli in the images. The eventual implementation of the system will be semi-automatic, where the best candidate images containing bacilli are presented to the medical technologist together with a bacillus count, confidence measures and recommended diagnosis. The final diagnosis could be performed by the technologist in less than a minute for typical cases. Furthermore, the results should be more accurate due to the higher number of view-fields processed. The study presented in this paper indicates that machine-assisted diagnosis of tuberculosis is certainly feasible.