Quantifying microscopic droplets in colloidal systems through machine learning-based image analysis

Different Machine Learning (ML) algorithms for object segmentation in microscopy images have been applied to various micrographs (acquired with different microscopy modalities) of different food colloidal systems. They are being compared to the manual segmentation of the droplets in each image and to traditional intensity thresholding segmentation. The ML algorithms showed varying levels of accuracy depending on the type of microscopy used and other characteristics of the images. In this work, an image analysis workflow has been developed. The workflow segments droplets of colloidal systems in microscopy images across a wide range of images and was specifically developed to quantify droplets with a large variation in size within the same image. Besides segmentation, the workflow also integrates the obtention of quantitative data from all droplets in the image. It is shown how the data can be used for different quantitative analyses regarding spatial distribution (droplet packing) and droplet shape. Finally, it is shown how this image analysis workflow segments and quantifies droplets inside other droplets, as in the case of double emulsions. The image analysis workflow, MIDAS, is made available as an open-source code in online repositories. The implementation of accurate image analysis will help bring new data to the fields of food colloids and food structure research. • Machine learning-based methods can segment microscopic colloidal droplets • The performance of the methods varies depending on the colloidal system and the images • An image analysis workflow for droplet segmentation and quantification is presented • The workflow works for droplets of varying sizes and in a broad range of images • New quantitative data becomes available for colloids and food microstructure research