TbImages: a smear microscopy image dataset to support the development of automated bacilli detection in the diagnosis of tuberculosis
In order to contribute to the development of automatic methods for the detection of bacilli, TBimages is an image dataset composed of two subsets: TbImages_SS1 contains 10 images per field, of different focal depths, and aims to support the definition of autofocus metrics and also the development of extended focus imaging methods to facilitate the detection of bacilli in smear microscopy imaging. TbImages_SS2 aims to support the development of automatic bacilli detection. Materials: Sputum smear microscopy slides from 12 patients prepared using the Kinyoun acid-fast stain and counterstained with methylene blue solution. The sputum samples are from patients suspected of pulmonary TB. Kinyoun stain was the method of staining used. It is similar to the Ziehl-Neelsen stain but does not require heating unlike the Ziehl–Neelsen stain. Image acquisition - An acquisition workstation was set up (a digital camera, a light microscope, and a computer). The digital camera used was a Canon PowerShot A640, which couples with a 10-megapixel CCD imager sensor with a 4x optical zoom. The images have 2816 x 2112 pixels and 24 bits per pixel (RGB images). The microscope used was a Zeiss Axioshop 40. It employed a magnification of 100x and numerical aperture of 1.25. TbImages_SS1: This subset contains 40 stacks of 10 microscope field images where the images of the stack were acquired at different focal depths. The fifth image in a stack is the one with the best focus. This dataset contains 400 images. TbImages_SS2: The images from this dataset were taken from 12 slides, 10 fields per slide, totaling 120 images. Annotated images (Gold Standard): In all 120 images of the TbImage_SS2 subset, objects of interest were identified and delimited within a geometric shape by a trained technician. A true bacillus was circled, an agglomerated bacillus was surrounded by a rectangle and an object whose image focus or geometry did not allow a clear identification as a bacillus (doubtful bacillus ) was surrounded by a polygon. In both databases, according to the background content, and density of bacilli in the images have been grouped as belonging to high background content or low background content, and high and low bacilli density, respectively.
More information: www.tbimages.ufam.edu.br
.Instructions are contained in the file New_readme_TbImages
- TbImages_subset 2 - bacilli detection evaluation. TbImages_SS2_ bacilli-detection-evaluation-20221129T095227Z-001.zip (545.17 MB)
- TbImages_subset 1 - autofocus evaluation. TbImages_SS1_autofocus evaluation-20221129T095751Z-001.zip (1.44 GB)