No abstract

Because of their widespread diffusion and impact on human health, early identification of pathogens responsible for urinary tract infections (UTI) is one of the main challenges of clinical microbiology. Currently, bacteria culturing on Chromogenic plates is widely adopted for UTI detection for its readily interpretable visual outcomes. However, the search of alternate solutions can be highly attractive, especially in the rapidly developing context of bacteriology laboratory automation and digitization, as long as they can improve cost-effectiveness or allow early discrimination. In this work, we consider and develop hyperspectral image acquisition and analysis solutions to verify the feasibility of a "virtual chromogenic agar" approach, based on the acquisition of spectral signatures from bacterial colonies growing on blood agar plates, and their interpretation by means of machine learning solutions. We implemented and tested two classification approaches (PCA+SVM and RSIMCA) that evidenced good capability to discriminate among five selected UTI bacteria. For its better performance, robustness and attitude to work with an expanding set of pathogens, we conclude that the RSIMCA-based approach is worth to be further investigated in a clinical usage perspective.

Abstract. The rapidly increasing diffusion of Full Microbiology Laboratory Automation plants is reshaping the way microbiologists perform diagnostic tasks. A huge stream of digital visual data is expected to be produced daily in the coming years in the emerging field of Digital Microbiology Imaging. In this context, we want to assess the suitability and effectiveness of a Deep Learning approach to solve the diagnostically relevant but visually challenging task of directly identifying pathogens on bacterial growing plates. In particular, starting from hyperspectral acquisitions in the VNIR range and spatial-spectral processing of cultured plates, we approach the identification problem as the classification of computed spectral signatures of the bacterial colonies. In a highly relevant clinical context (urinary tract infections) and on a database of HSI images, we designed and trained a Convolutional Neural Network for pathogen identification, assessing its performance and comparing it against conventional classification solutions.