Segmentation and classification of bacterial culture images

Abstract

A procedure using simple image processing and pattern recognition techniques to automatically extract and count organisms using images of bacterial cultures is described in this paper. Methanospirillum hungatei and Methanosarcina mazei were used for the experiments. The methanogens were grown in anaerobic serum vials for 1–3 week period. Samples of the growing cultures were withdrawn from the vials and photographed. The procedure was applied to images containing one kind of organism as well as images containing both type of organisms. The procedure separated and correctly identified the cultures as M. hungatei and M. mazei even in images where the organisms were overlapping or touching. The edge detection, the splitting, and the classification stages, by using this procedure, were accomplished in less than a minute. The image analysis and pattern recognition techniques described in this paper are simple, fast, and able to identify organisms based on their shapes. Our results indicate that such systems can be used to semi-automate routine culture-automate routine culture methods allowing rapid direct microscopic techniques for the detection, enumeration, and identification of bacteria.