Automatic Computational Method for Viable Cell Counting on Microphotographs of Mature Biofilms

Objectives: The aim of this research was to develop a computational algorithm for quantitative assays that automatically detects and counts cells disrupted from a mature biofilm.
Methods: Single species biofilms of Prevotella intermedia, Actinomyces viscosus, Aggregatibacter actinomycetemcomitans and Streotoccoccus sanguinis, were grown on cellulose nitrate membrane filters (0.2 μm pore size, 5 mm diameter; Whatman International Ltd, Maidstone, UK). Mean cell counts of the disrupted biofilms were performed using a viability stain (Live/Dead BacLight; Invitrogen, Paisley, UK). Five-point fields were selected at random for image acquisition and cell counting (viable and non viable cells) using an epifluorescence microscope (Leica, X 20 objective). An image of the fields with dyed bacteria was taken on grayscale image for red and green colours. Merging of the images (RGB) was performed by employing ImageJ. The observable of interest was the ratio of live-to-total and dead-to-total cells. The algorithm was developed in Matlab and it analysed the input RGB image, calculated the coordinates of cells in the image and counted them. The main criteria for analysing the image were the object size and brightness.
Results: The output of the automated method resembled very closely the counts obtained by direct visual inspection of the images. For most cases the difference between visual and automated counting were of order 5%. There were some cases where the difference could be bigger -up to 30%, mostly due to unidentified, unusual artifacts in the image. However, the automated method seemed to always calculate correctly which colour is dominating.
Conclusions: The method seemed to reproduce the cell counts ratios sufficiently well. An assumption could be made about maximum local density of found cells. Regions with abnormally high density could be identified as artifacts and removed.