DEVELOPMENT OF ARTIFICIAL NEURAL NETWORKS CAPABLE OF IDENTIFYING CRYPTOSPORIDIUM PARVUM OOCYSTS STAINED WITH 4′, 6 DIAMIDINO‐2‐PHENYLINDOLE

Abstract

Cryptosporidium parvum is a coccidian protozoan parasite capable of infecting a variety of mammalian hosts, and can cause gastro‐enteric disease within humans. C. parvum oocysts were stained with varying concentrations of 4′, 6 diamidino‐2‐phenylindole (DAPI). After microscopic observation, objects of interest were captured using a CCD color digital camera. The microscopic images were classified based on their DAPI stain properties as either DAPI positive or negative. Individual oocysts were cropped, converted to grayscale, applied to a binary threshold filter, and were further processed into a numerical data array. DAPI positive and negative images (100 each) were randomly removed for artificial neural network (ANN) testing. The remaining image data were used for ANN training using a commercially available software program. After training experimentation, a final network design was implemented possessing 95 input, 400 hidden, and 2 output neurons. Additional control image sets (ranging from 165 to 119 images) were collected to better ascertain ANN performance. These images consisted of DAPI positive oocysts and two types of DAPI negative images (either formalin treated oocysts or algal cultures). Selected ANN correctly identified, as a range, 82.4–93.8% of the DAPI positive oocysts, 97–98.2% of the DAPI negative oocysts, and 52.9–57% of the DAPI negative algal cells. The control image sets were unique data, never presented during ANN training. Because of this, combined with the high number of correct image identifications for certain image sets, ANN technology may provide a means to identify C. parvum oocysts through automated analysis.