Characterizing temporal development of biofilm porosity using artificial neural networks

Abstract

We used artificial neural networks (ANN) to compute parameters characterising biofilm structure from biofilm images and to interpolate a limited number of experimental data characterising the effects of nutrient concentration and flow velocity on the areal porosity of biofilms. ANN were trained using a set of experimental data characterising structural parameters of biofilms of Pseudomonas aeruginosa (ATCC #700829), Pseudomonas fluorescens (ATCC #700830) and Klebsiella pneumoniae (ATCC #700831) for various flow velocities and glucose concentrations. We used 80% of the data to train ANN and 10% of the data to validate the results, which is routinely carried out as a countermeasure against overtraining. Trained ANN were used to interpolate into the data set and evaluate the missing 10% of the data. To compare ANN accuracy in evaluating the missing data with the accuracies achieved using other interpolation algorithms, we used spline, cubic, linear and nearest-neighbour interpolation algorithms to evaluate the missing data. ANN estimates were consistently closer to the experimental data than the estimates made using the other methods.