Identification of tuberculosis bacillus in direct smears with low image quality

摘要

A digital image processing method for detecting tuberculosis (TB) is presented. First, a reference representing the specialists' determination of TB bacilli on a digital image is validated. And the noise in the signal intensities is evaluated to quantify the image capture system. A simple algorithm is proposed for identification of the bacillus, by means of only thresholding in the color channels. This simple method causes a sensitivity of 86.08% and specificity of 92.39% in classifying the pixels of poor quality images. Although, there are methods with a better sensitivity and specificity for the detection of bacillus, these have often more complex processes and use input images with a better contrast. Our results show a potential to obtain an automatic detection of TB bacillus.