Avtomatska segmentacija celičnih predelkov v volumetričnih podatkih pridobljenih z elektronskim mikroskopom

In recent years, electron microscopy has enabled the acquisition of volumetric data with resolving power to directly observe the ultrastructure of intracellular compartments. New insights and knowledge about cellular processes offered by such data require comprehensive analysis, which is limited by the time-consuming manual segmentation and reconstruction methods. We present methods for automatic segmentation, reconstruction and analysis of intracellular compartments from volumetric data obtained by dual-beam electron microscopy which combines a focused ion beam and a scanning electron microscope (FIB-SEM). In particular, we focus on the segmentation of mitochondria, endolysosomes, fusiform vesicles and the Golgi apparatus, the reconstruction of mitochondria and fusiform vesicles and the morphological analysis of the reconstructed mitochondria. The segmentation methods are based on supervised deep learning and include mechanisms that reduce the impact of dependencies in the input data, artifacts and annotation errors, while the reconstruction methods are based on more traditional image processing techniques that incorporate knowledge about the morphology of the structures. We present UroCell, a new publicly available volumetric electron microscopy (EM) dataset on which we evaluated our methods. The evaluation revealed that the proposed methods for segmentation, reconstruction and analysis achieve higher accuracy compared to existing state-of-the-art methods.