Ayse Aydogdu-Erozan, Philipp D. Lösel, Vincent Heuveline, Venera Weinhardt
The structure of cells is a key to understanding cellular function, diagnosis of pathological conditions, and development of new treatments. Soft X-ray tomography (SXT) is a unique tool to image cellular structure without fixation or labeling at high spatial resolution and throughput. Ongoing improvements in faster acquisition times increase demand for accelerated image analysis. Currently, the automatic segmentation of cellular structures is a major bottleneck in the SXT data analysis pipeline. In this paper, we introduce an automated 3D cytoplasm segmentation model – ACSeg – by use of semi-automatically segmented labels and 3D U-Net, implemented in the online platform Biomedisa. The segmentation model is trained on semi-automatically labeled datasets and shows rapid convergence to high-accuracy segmentation, therefore reducing time and labor. ACSeg trained on 43 SXT tomograms of human immune T cells, the model successfully segmented unseen SXT tomograms of human hepatocyte-derived carcinoma cells, mouse microglia, and embryonic fibroblast cells. Furthermore, we could diversify the model by adding only 6 specific SXT tomograms, showing the potential for the development of an optimal experimental design. The ACSeg is published on the open image segmentation platform Biomedisa, enabling high-throughput analysis of cell volume and structure of cytoplasm in diverse cell types. The approach can be expanded for automatic segmentation of other organelles visualized by SXT, providing means for structural analysis of cell remodeling under different pathogens at statistically significant sizes, therefore enabling the development of novel drug treatments.