Automated image processing of fluorescent images: the end to manual counting?

Abstract

High-throughput experiments are characterized by the generation of large amounts of imaging data in short time which requires fast, automated and reliable methods for processing and analysis. Most of the gold standard meth­ods for quantitative and semi-quantitative analysis of fluorescent images are relying on human decision-making. The existing automated analysis methods require very strict maintenance of constancy in all experimental condi­tions from animal care and material acquisition to image acquisition. In the high-throughput experiments, how­ever, even the smallest biases introduced by human involvement and the tiniest fluctuations of experimental con­ditions may be multiplied to unacceptable magnitudes. Thus, any image processing and analysis algorithm, part of such an experiment, must reduce human-produced biases and be robust to variability in experimental condi­tions. We propose a processing sequence suitable for the analysis of images produced by wide-field epifluores­cence RNA-FISH and FIHC experiments which is easily adjustable to the conditions of different experiments. The algorithm involves the following steps: (i) Optimal image acquisition of multichannel z-stack images; (ii) Shad­ing correction of the multidimensional images; (iii) Single manifold extraction of an accurate high-contrast de­blurred 2D-projection; (iv) Correction of exposure with emphasis on background normalization in the final im­age. With this algorithm, we produced images for automated classification of cells expressing different markers. After the initial processing a machine learning algorithm segments the DAPI stained channel and perinuclear re­gions of interest are produced. Next, the marker channels overlap with the perinuclear RoIs was used to classify each cell into positive or negative for the corresponding marker. The proposed sequence of image processing and analysis steps is performed with an easily procurable open source software (Fiji/ImageJ and Ilastik) which allows full automation thanks to its scripting capabilities.