Colony Counting with Artificial Intelligence

Introduction: Culture-based methods and colony enumeration are central in microbiology for quality assessment in food, water and pharmaceutical analysis to ensure public health. However, manual enumeration is time-consuming, labour-intensive and subjective.

Objective: To overcome the above-mentioned limitations, we have set ourselves the goal of developing user-friendly, adaptive image processing methods by using artificial intelligence (AI) tools.

Materials & Methods: A multi-layered strategy that integrated pre-trained and novel convolutional neural networks (CNNs) was employed by including data acquisition, preparation and model training.

Results: By using a self-developed FHOTOBOX, we compiled a data set of colony images by using agar plates of diverse media showing bacterial colonies from food sample analyses. Our innovative deep learning approach led to successful segmentation of individual colonies with high accuracy, overcoming classic computer vision hurdles such as overlap and boundary errors. This strategy also allowed us to effectively process multiple image sources, demonstrating the adaptability and efficiency of deep learning. Based on the pre-trained CNNs and transfer learning, we developed a web application that can be used on a tablet for fast and reliable enumeration of bacterial colonies.

Conclusions: AI systems, especially in image processing, can facilitate time-consuming routine procedures such a bacterial colony counting in microbiologiy laboratories. It is conceivable that the technology could one day be integrated as a low-cost mobile app.