Quantification of golgi dispersal and classification using machine learning models

The Golgi body is a critical organelle in eukaryotic cells responsible for processing and modifying proteins and lipids. Under certain conditions, such as stress, disease, or ageing, the Golgi structure alters. Therefore, understanding the mechanisms that regulate Golgi dispersion has significant research contributions to identifying disease. However, there is a lack of tools to quantify the Golgi dispersion datasets. In this paper, we aim to automate the process of quantification of Golgi dispersion and use extracted features to classify dispersed Golgi images from undispersed Golgi images using machine learning models. First, we collected confocal microscopy images of transiently transfected HeLa cells expressing Galactose-1-phosphate uridylyltransferase (GALT)- green fluorescent protein (GFP) to quantify Golgi dispersal and classification. For the quantification, we introduced automated image processing and segmentation by applying mean and Gaussian filters. Then we used Otsu thresholding on preprocessed images and watershed segmentation to refine the segmentation of dispersed Golgi particles. In the case of classification, we extracted features from the Golgi dispersal images and classified them into empty vector (EV) versus CARP1 ring mutant (CARP1 RM) and empty vector (EV) versus CARP1 wildtype (CARP1 WT) classes. Our approach used machine-learning models, including logistic regression, decision tree, random forest, Naive Bayes, k-Nearest Neighbor (KNN), and gradient boosting for dispersed Golgi image classification. The experiment results show that our quantification technique on Golgi dispersal images reached 65% classification accuracy when the system uses a gradient boosting classifier for EV vs. CARP1 WT classification. Furthermore, our approach achieved 65% classification accuracy using a random forest classifier for EV vs. CARP1 RM classification.