Automatic segmentation of intracellular compartments in volumetric electron microscopy data

Abstract

Segmentation of intracellular compartments is a technique that provides quantitative data about the presence, spatial distribution, structure and consequently the function of intracellular compartments, the central organization units of eukaryotic cells. With the recent development of high throughput data acquisition techniques in electron microscopy, manual segmentation is becoming a major bottleneck of the process. To aid biomedical research, we propose a technique for the automatic segmentation of mitochondria and compartments of the lysosomal pathway from cells obtained from the mammalian bladder with the focused ion beam combined with the scanning electron microscopy technique (FIB-SEM). We propose a segmentation pipeline based on the convolutional neural network that exploits the fact that mitochondria and compartments of the lysosomal pathway have similar textural features in certain regions. Using this knowledge, our approach outperforms existing state-of-the-art models evaluated on our dataset.