MitoNet and empanada

Empanada (EM Panoptic Any Dimension Annotation) is a tool for panoptic segmentation of organelles in electron microscopy (EM) images in 2D and 3D. Panoptic segmentation combines both instance and semantic segmentation enabling a holistic approach to image annotation with a single deep-learning model. To make model training and inference lightweight and broadly applicable to many EM imaging modalities, Empanada runs all expensive operations on 2D images or run-length encoded versions of 3D volumes.

Manuscript on BioRxiv:

CEM1.5M

A heterogeneous, non-redundant, information-rich, and relevant unlabeled EM image dataset (at ∼1.5 x 106 images, the largest of its kind to our knowledge) for use as a database to pre-train and sample images for mitochondria, or other organelles, segmentation. We used CEM1.5M to pre-train MitoNet

Dataset uploaded at:

CEM-MitoLab

A dataset of ~22K cellular EM 2D images with label maps of ~135K mitochondrial instances, for deep learning. We used this labeled dataset to train MitoNet.

Dataset uploaded at:

MitoNet Benchmark

7 benchmark datasets with instance labeled mitochondria, used to test MitoNet. 1 dataset is a series or random TEM images; the others are isotropic volume EM (FIB-SEM) data from various biosamples and sample preparation techniques.

Dataset uploaded at:

CEM500K

A large-scale heterogeneous unlabeled cellular electron microscopy image dataset for deep learning methods.

Dataset uploaded at:

Full Code Uploaded:

Manuscript on BioRxiv:

Cryo LM of high-pressure-frozen thick samples (C. elegans)

This video describes a workflow to execute cryo fluorescence microscopy on high-pressure-frozen C. elegans worms. This collaboration with Carl Zeiss GmbH was presented (Poster #1085) at Microscopy & Microanalysis 2020.

“2.5D” Neural Network based segmentation of mitochondria

This video describes a method to enforce prediction consistency across orthogonal planes (hence 2.5D) to improve neural network segmentation of mitochondria in a FIB-SEM image volume. This work was presented (poster #831) at Microscopy & MIcroanalysis 2020.

Towards artifact-free FIB-SEM acquisition

This video describes artifact-free FIB-SEM data acquisition strategies, and accompanied a poster (#535) presented at Microscopy & Microanalysis 2020

Editing mitochondrial skeletons in a virtual reality (VR) environment

In this movie, we describe a module in ArivisVR to import, edit and export corrected mitochondrial skeletons. This work (in collaboration with arivis Inc) was presented (Poster #448) at Microscopy & Microanalysis 2020.