Segmentation of three-dimensional (3D) electron microscopy (EM) image stacks is an arduous and tedious task. Deep convolutional neural networks (CNNs) work well to automate the segmentation; however, they require a large training dataset, which is a major impediment. In order to solve this issue, especially for sparse segmentation, we used a CNN with a minimal training dataset. We segmented a Cerebellar Purkinje cell from an image stack of a mouse Cerebellum cortex in less than two working days, which is much shorter than that of the conventional method. We concluded that we can reduce the total labor time for the sparse segmentation by reducing the training dataset.
Three-dimensional (3D) observation of a biological sample using serial-section electron microscopy is widely used. However, organelle segmentation requires a significant amount of manual time. Therefore, several studies have been conducted to improve their efficiency. One such promising method is 3D deep learning (DL), which is highly accurate. However, the creation of training data for 3D DL still requires manual time and effort. In this study, we developed a highly efficient integrated image segmentation tool that includes stepwise DL with manual correction. The tool has four functions: efficient tracers for annotation, model training/inference for organelle segmentation using a lightweight convolutional neural network, efficient proofreading, and model refinement. We applied this tool to increase the training data step by step (stepwise annotation method) to segment the mitochondria in the cells of the cerebral cortex. We found that the stepwise annotation method reduced the manual operation time by one-third compared with that of the fully manual method, where all the training data were created manually. Moreover, we demonstrated that the F1 score, the metric of segmentation accuracy, was 0.9 by training the 3D DL model with these training data. The stepwise annotation method using this tool and the 3D DL model improved the segmentation efficiency for various organelles.
Array tomography (AT) provides three-dimensional (3D) information by observing serial sections using scanning electron microscopy (SEM). Compared with serial block-face SEM and focused ion beam (FIB)/SEM, AT has several advantages: (a) High lateral resolution, (b) Standard staining with uranium and lead, allowing easy comparison of images captured by SEM and transmission electron microscopy (TEM), (c) Repetitive observations of samples, enabling hierarchical analysis from low magnification to high magnification, and (d) Low installation cost. In many cases, however, images of serial sections have been taken manually with labor and time. In addition, manual segmentation has been usually required. To solve these issues, we developed automatic image capturing software for AT with SEM. We also tried to use deep neural network to assist segmentation of the images.In Figure 1, flow diagram of observation and analysis is shown. Mouse cerebellum cortex was fixed with glutaraldehyde and osmium tetra oxide, and was embedded in epoxy resin. Then serial sections were cut using ultra microtome and scooped on silicon substrate, and they were stained with uranyl acetate and lead citrate. SEM (JSM-7800F and JSM-7900F; JEOL Ltd.) was used for capturing images.A software was developed for automatic capturing of SEM images in each serial section at the same position following manual identification of the corresponding apex of serial sections. Convolutional neural network algorithm was used to extract cell nuclei in each image. Since such an automatic segmentation was not perfectly completed, we manually corrected the data after the automatic segmentation. Then, 3D image was reconstructed by the stack of images from serial sections, and evaluated with quantitative analysis.In Figure 2, images of cell nuclei in various cells are shown. Since the contrast of images was similar to that obtained by TEM, microstructures including cell nuclei were easily identified. A 3D-reconstructed image is shown in Figure 3. The automatic image capturing software and deep neural network analysis brought about easier procedure for the segmentation and quantitative analysis. Such a system for analysis of SEM images with AT will realize to be used for pathological diagnosis as well as basic biology.References:[1] Micheva et. al, Neuron, 55 (2007), p. 25.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.