Creating and sharing knowledge for telecommunications

Determining Leishmania Infection Levels by Automatic Analysis of Microscopy Images

Nogueira, Pedro A. Nogueira ; Coimbra, M.

Determining Leishmania Infection Levels by Automatic Analysis of Microscopy Images, Proc Congresso de Métodos Numéricos em Engenharia, Coimbra, Portugal, Vol. ., pp. . - ., June, 2011.

Digital Object Identifier:

Manual counting of microscopy images is both tiring and unreliable for biologists. In this paper we address this by automatically determining Leishmania infection levels in microscopy images using computer vision methodologies. For this purpose 794 fluorescence microscopy images were collected and used for this study. All of them were manually annotated by biologists using our CellNote [1] software platform.
Images are pre-processed by contrast stretching (to normalize illumination conditions) and by low-pass filtering (for noise reduction). Three algorithms are applied separately to each image. Algorithm A intends to detect cell nuclei and consists of a segmentation via adaptive multi-threshold, and classification of resulting regions using a set of collected features (area, centre of mass, shape) into either cell nucleus, set of cell nuclei, or other. Algorithm B intends to detect cell parasites and is similar to Algorithm A but with a modified version of the previous adaptive multi-threshold and each region is classified into either parasite, set of parasites, or other. Algorithm C intends to detect the cytoplasm of cells and parasites and consists of a shorter version of the previous two algorithms, were no classification is made. We addressed the regions with multiple nuclei or parasites using Gaussian mixture models. The study was concluded by associating each parasite to a single cell using minimum Euclidean distance to a cell’s nucleus, thus quantifying Leishmania infection levels. We were able to count cells and parasites with reasonably high accuracies (above 90%), and decluster regions with multiple nuclei or parasites with 75-80% accuracy, providing biologists with a useful tool for Leishmania research.