From Segmentation to Clustering: Advancing Algorithms for Grouping Microbial Colonies on Agar Plates

Agar plates are essential in microbiology, yet manual intervention remains common for tasks like colony-picking, especially in smaller laboratories. This paper addresses the research gap of clustering microbes within agar plate images to advance automated solutions. Three innovative clustering approaches based on visual properties are presented. Using synthetic agar plate images with annotated microbes, these algorithms are developed and evaluated, with performance assessed using the V-measure metric. Results reveal challenges in pixel-level clustering without segmentation masks (V-measure: 0.284), while U-Net autoencoder features with OPTICS show reasonable performance (V-measure: 0.705). K-means with a colony extraction pipeline excels (V-measure: 0.942), emphasizing the importance of high-level colony features. These solutions offer potential enhancements for colony-picking robots and advanced analysis. This research significantly contributes to agar plate analysis, paving the way for future automated colony analysis and streamlined lab workflows.