An increasing incidence of device-related, biofilm-associated infections has been observed in clinical practice worldwide. In vitro biofilm models are essential to study these burdensome infections and to design and test potential new treatment approaches. However, there is considerable variation in in vitro biofilm models, and a generally accepted systematic description of biofilm maturity - apart from incubation time - is lacking. Therefore, we proposed a scheme comprised of 6 different classes based on common topographic characteristics, i.e., the substrate, bacterial cells and extracellular matrix, identified by atomic force microscopy (AFM), to describe biofilm maturity independent of incubation time. Evaluation of a test set of staphylococcal biofilm images by a group of independent researchers showed that human observers were capable of classifying images with a mean accuracy of 0.77 ± 0.18. However, manual evaluation of AFM biofilm images is time-consuming, and subject to observer bias. To circumvent these disadvantages, a machine learning algorithm was designed and developed to aid in classification of biofilm images. The designed algorithm was capable of identifying pre-set characteristics of biofilms and able to discriminate between the six different classes in the proposed framework. Compared to the established ground truth, the mean accuracy of the developed algorithm amounted to 0.66 ± 0.06 with comparable recall, and off-by-one accuracy of 0.91 ± 0.05. This algorithm, which classifies AFM images of biofilms, has been made available as an open access desktop tool.

This paper explores a multicriteria spatial analysis methodology with a machine learning algorithm, the Classification Tree Analysis (CTA) within Idrisi GIS, to classify and identify homogeneous regions. The proposed approach is tested in a case study carried out in the South of Brazil. All the municipalities were classified and grouped within areas according to similar condition of urban preponderance in socioeconomic and environmental indicators. The results are evaluated and compared with two other methodologies previously implement by the authors: (a) a ranking of municipality through an aggregate index; and (b) using Kohonen´s Self-Organizing Map (SOM) as an unsupervised classifier. The identification of similar areas with analogous socioeconomic and environmental characteristics is important to the development of regional and municipal common sustainable strategies and advances in municipality partnerships.