Detection and Counting of Lipid Droplets in Adipocyte Differentiation of Bone Marrow-Derived Mesenchymal Stem Cells Using a Tiny Convolutional Network and Image Processing

Abstract

Stem cells are a bunch of cells that are considered as encouraging cells for treating patients because of their ability to regenerate themselves and also their potential for differentiation into several lineages. When stem cells are differentiated into adipose tissues, a great variety of lipid droplets usually grow in these cells and can be observed by oil red O staining, which is typically used for evaluating adipocyte differentiation status. For numerous differentiation experiments, counting and calculation of the population of lipid droplets are necessary. The disadvantages of conducting experiments for identification and investigation of lipid droplets include being expensive, time-consuming and subjective. There are few studies carried out in the field of machine learning and image processing for the automatic detection and counting of lipid droplets in intracellular images. In this study, to demonstrate the adipocyte differentiation of mesenchymal stem cells, their microscopic images were prepared. After the preprocessing operation, the images were fed to a tiny convolutional neural network. Images created within the network output were examined using two image processing methods. Finally, the number of lipid droplets was obtained with acceptable accuracy, and their exact location was displayed.