Wellcounter: Automated high‐throughput phenotyping for aquatic microinvertebrates

Abstract Using phenotypic information extracted from digital images through automated computer vision analysis has become a promising avenue in ecological and evolutionary research. However, acquiring large quantities of high‐quality image data and processing them efficiently often remain a bottleneck. Here, we introduce the Wellcounter, an advanced platform designed for the automated high‐throughput phenotyping of aquatic microinvertebrates (100–2000 μm), utilizing common multiwell plates to facilitate large‐scale ecological experiments. The system automates image acquisition, processing and analysis, enabling the tracking of hundreds of populations with minimal daily manual effort. Key hardware components include a high‐resolution digital camera, a telecentric lens and a motorized XY linear guide system based on an open‐source design that moves to each individual well. A dark‐field illumination ring, which travels synchronously with the camera and lens, provides constant and standardized light conditions and enhances the visualization of microinvertebrates within each well, improving detection accuracy. Our prototype can handle 84 six‐well plates in two batches, ensuring efficient imaging and analysis of over 500 populations. The software, developed in Python, supports detailed population size measurements, growth rate analysis and swimming behaviour quantification. It includes modules for image acquisition control, image analysis and motion analysis, ensuring comprehensive data collection and interpretation. We also provide a module for generating ‘ground truth’ datasets, allowing to label the positions of all detectable specimens in an image, and an optimization module to fine‐tune imaging parameters for a new study organism. Validation of the Wellcounter was conducted using rotifers. The platform demonstrated high accuracy in detecting and counting individuals, as well as in measuring population growth rates and swimming behaviour. Image analysis parameters were optimized to minimize false negatives and false positives. We discuss adapting this protocol for different microorganisms and hardware configurations, as well as general caveats and current limitations of this phenotyping approach. For those interested in developing image analysis algorithms, we provide large annotated datasets, including high‐resolution movies and images with known quantities and positions of specimens. The approach and resources provided in this study aim to facilitate the adoption and further development of automated image analysis technologies in ecological research.