RAPID-DETERMINATION OF BACTERIAL ABUNDANCE, BIOVOLUME, MORPHOLOGY, AND GROWTH BY NEURAL-NETWORK-BASED IMAGE-ANALYSIS

Annual bacterial plankton dynamics at several depths and locations in the Baltic Sea were studied by image analysis. Individual bacteria wer e classified by using an artificial neural network which also effectiv ely identified nonbacterial objects, Cell counts and frequencies of di viding cells were determined, and the data obtained agreed well with v isual observations and previously published values. Cell volumes were measured accurately by comparison with bead standards. The survey incl uded 690 images from a total of 138 samples. Each image contained appr oximately 200 bacteria. The images were analyzed automatically at a ra te of 100 images per h, Bacterial abundance exhibited coherent pattern s with time and depth, and there were distinct subsurface peaks in the summer months. Four distinct morphological classes were resolved by t he image analyzer, and the dynamics of each could be visualized. The b acterial growth rates estimated from frequencies of dividing cells wer e different from the bacterial growth rates estimated by the thymidine incorporation method. With minor modifications, the image analysis te chnique described here can be used to analyze other planktonic classes .