Segmentation of Petri Plate Images for Automatic Reporting of Urine Culture Tests

Recently, significant improvements in biological and medical decision support systems have been obtained by using hybrid methods, based on a combination of advanced image processing techniques, artificial intelligence tools, fuzzy logic, genetic algorithms, and Bayesian modeling. In particular, the development of intelligent tools for the automatic reporting of medical analyses (screening systems) has attracted increasing research interest, due to their higher reliability, accuracy, reduced staff time, and lower costs. In this chapter, we propose a survey of computer vision and machine learning methods employed for the urine culture screening based on Petri plate automatic image understanding. Petri plates are used for bacterial cultures, which are employed in a wide variety of microbiological tests, from food and beverage safety assessments to environmental control, and to many specific clinical analyses (f.i. urine culture). Several segmentation techniques and some specific approaches to perform bacterial counting and infection classification are described below, along with a synthetic image generation approach required to overcome privacy concerns and medical data paucity. Indeed, during the last decade, deep learning has had a devastating impact on image processing, achieving exceptional results. Nonetheless, most of these improvements rely on fully annotated data, being the annotation procedure inherently difficult and expensive. The generated synthetic annotated images can be profitably used to train deep architectures, enabling reliable image segmentation.