Automatic Reconstruction of Mitochondria and Endoplasmic Reticulum in Electron Microscopy Volumes by Deep Learning

Abstract

Together, mitochondria and the endoplasmic reticulum (ER) occupy more than 20% of a cell's volume, and morphological abnormality may lead to cellular function disorders. With the rapid development of large-scale electron microscopy (EM), manual contouring and three-dimensional (3D) reconstruction of these organelles has previously been accomplished in biological studies. However, manual segmentation of mitochondria and ER from EM images is time consuming and thus unable to meet the demands of large data analysis. Here, we propose an automated pipeline for mitochondrial and ER reconstruction, including the mitochondrial and ER contact sites (MAMs). We propose a novel recurrent neural network to detect and segment mitochondria and a fully residual convolutional network to reconstruct the ER. Based on the sparse distribution of synapses, we use mitochondrial context information to rectify the local misleading results and obtain 3D mitochondrial reconstructions. The experimental results demonstrate that the proposed method achieves state-of-the-art performance.

Interorganelle communication, aging, and neurodegeneration

“…More recently, DJ-1 has also been reported to interact with the Grp75-VDAC-IP3R tethering complex to stabilize MERCs and facilitate Ca 2+ transfer. In the absence of DJ-1, Grp75, VDAC1, and IP3R interact less, resulting in smaller MERCs, decreased mitochondrial Ca 2+ uptake, and decreased ATP production (Liu et al 2020). Overexpression of DJ-1 can also enhance mitochondrial Ca 2+ uptake (Ottolini et al 2013).…”

“…Advances in various technologies (Scorrano et al 2019;Huang et al 2020) are essential to be able to visualize, identify, and functionally evaluate the dynamic interplay between distinct contact sites and provide greater understanding of the cellular communicome. Electron microscopy approaches like FIB-SEM (Xu et al 2019), and automated reconstruction with technologies like deep learning (Liu et al 2020), are enabling direct visualization of cellular organelles and providing an in-depth view of contact sites. Applying such techniques to cells from patients or cells with specific engineered disease mutations could enable unbiased global analysis of perturbed contact sites in various pathologies, identifying the potentially relevant sites.…”

Deep Learning-Based Classification of GAD67-Positive Neurons Without the Immunosignal

“…For example, the lower (e.g., the first and the second) layers extract geometric features such as edges and their arrangement in images, whereas the higher (e.g., the penultimate and the last) layers identify abstract concepts and complicated characteristics. Deep learning-based image processing methods have recently improved and been applied in biological fields (He et al, 2015 ; Jiménez and Racoceanu, 2019 ; Kusumoto and Yuasa, 2019 ; Liu et al, 2020 ).…”

An interactive deep learning-based approach reveals mitochondrial cristae topologies

“…Particularly, variants of U-Net have been used successfully for segmentation of biological images (Falk et al, 2019). The segmentation of mitochondria using deep learning has improved recently (Liu et al, 2020;Xiao et al, 2018;Žerovnik Mekuč et al, 2020). While image calculation speed increased, laborious manual processing, such as generation of training data, proofreading of the analysis, and converting files for processing images across multiple software programs is still a significant limitation in EM analyses.…”