An Integrated Approach for Coccidiosis Diagnosis through Digital Image Processing

This work reports on the development of a computer vision system for automated analysis of parasite stages by using state-of-the-art concepts and methods, that allows enhanced objectivity, speed, accuracy, reproducibility and documentability when compared to traditionally performed human visual inspection and diagnosis [1]. Seven distinct Eimeria species can infect chickens, causing intestinal lesions of variable extent and severity. Because different species can vary in pathogenicity, the precise discrimination is important for epizootiological studies. Oocysts of distinct species present differences of size, shape, thickness and color of the oocyst wall. However, the correct species assignment by visual inspection is severely restricted by the slight differences and overlap of characteristics among the different species. The currently considered geometrical representations and features to be used for characterization of the cell outline include but are not limited to: area, perimeter, elongation, tangent and normal fields estimated by spectral methods, number of singularities (e.g. inflection points), multiscale curvature [2] and multiscale bending energy [2]. The measures being used to characterize the nuclear chromatin include the multiscale fractal dimension [3], co-occurrency matrices and entropy. Data mining, feature selection, and statistical multivariate analysis, as well as variational approaches, should be used in order to try to identify possible new features that could be particularly effective for the considered problem. A distinctive characteristic of our project is that large, almost unlimited, amount of experimental data (high contrast and high quality images of Eimeria oocysts) can be produced under controlled conditions from the seven Eimeria species and organized into databases. Such an uncommon availability of controlled and high-quality data can be explored in order to construct complete statistical models leading to nearoptimal supervised classification through Bayesian approaches [2]. The selected techniques and classification methods should be incorporated into a graphic-interactive software application that could be used off-line or remotely through WWW. Another interesting perspective is to integrate and complement the above described morphological approach, comparing it to PCR-based assays already available [4]. It is also expected that such applicative will be adapted for the analysis and diagnosis of other important parasitic pathogens.