Multistage classification for bacterial colonies recognition on solid agar images

The advent of laboratory automation in clinical microbiology is entailing a revolution in the way most common bacteriological clinical exams are accomplished. As an essential part of these systems, digital recording and processing of cultured bacteria images is expected to improve plate reading, with a key role of image analysis tools in guaranteeing cost-effectiveness, accuracy, flexibility and reliability of the clinical tasks. In this work, we propose an image analysis system capable to address the complex problem of different bacteria species identification on cultured agar plates. Our solution is based on a modular segmentation/classification pipeline where a chain of supervised classification stages provides solutions to a series of nested task issues, from foreground separation toward isolated colony detection and classification. Performance assessment, based on an experimental dataset obtained in standardized laboratory conditions, clearly demonstrates the feasibility and the potentiality of the proposed solution and favorably opens to generalizations as well as to clinical validation studies.