Quantifying microscopic droplets in colloidal systems through machine learning-based image analysis
Collaboration
J. Saalbrink and J. C. Bonilla, Department of Green Technology, University of Southern Denmark
Research Background
Understanding the structure of food at microscopic scale is essential for improving food texture and stability. Food colloidal systems, like emulsions and suspensions, are mixtures where tiny droplets or particles are dispersed throughout a liquid. These droplets often vary greatly in size within the same sample, which makes it challenging for traditional segmentation tools to accurately identify them. Addressing this challenge, J. Saalbrink and J.C. Bonilla developed MIDAS (MIcroscopic Droplet AnalysiS), a Python-based workflow that extends the popular Cellpose segmentation algorithm to handle multiple droplet scales simultaneously, enabling precise segmentation and quantification of droplets across diverse microscopy images.
Method
MIDAS builds on Cellpose by performing segmentation multiple times with different size droplet settings predefined by the user. This multi-scale approach creates layered segmentation results that are later merged to eliminate duplicate detections of the same droplet across layers. The workflow incorporates unique labeling and morphological postprocessing steps to generate a final clean segmentation mask that captures droplets of all sizes. Beyond segmentation, MIDAS extracts quantitative data including the droplet size, spatial distribution (droplet packing), and shape features derived from contour analysis. Our facility supported the MIDAS project by reviewing and optimizing the Python code to enhance speed and efficiency, ensuring MIDAS can handle large datasets well.
For more details, see the full publication and code linked below.
Publication
Saalbrink, J., Loo, T. Y., Mertesdorf, J., Xu, P., Pedersen, M. T., Clausen, M. P., & Bonilla, J. C. (2025). Quantifying microscopic droplets in colloidal systems through machine learning-based image analysis. Food Hydrocolloids, 166, 111301. https://doi.org/10.1016/j.foodhyd.2025.111301