A semantic and instance segmentation toolbox for EM connectomics.
The field of connectomics aims to reconstruct the wiring diagram of the brain by mapping the neural connections at the level of individual synapses. Recent advances in electronic microscopy (EM) have enabled the collection of a large number of image stacks at nanometer resolution, but the annotation requires expertise and is super time-consuming. Here we provide a deep learning framework powered by PyTorch for automatic and semi-automatic image segmentation in connectomics.
MitoEM Dataset: Large-scale 3D Mitochondria Instance Segmentation
Electron microscopy (EM) allows identifying intracellular organelles such as mitochondria, providing insights for clinical and scientific studies. We introduce the MitoEM dataset, a 3D mitochondria instance segmentation dataset with two (30 µm)^3 volumes from human and rat cortices, respectively, 3, 600× larger than previous benchmarks. With around 40K instances, we find a great diversity of mitochondria in shape and density. On MitoEM, we find existing instance segmentation methods often fail to correctly segment mitochondria with complex shapes or close contacts with other instances.
AxonEM Dataset: 3D Axon Instance Segmentation of Brain Cortical Regions
Electron microscopy (EM) enables the reconstruction of neural circuits at the level of individual synapses, which has been transformative for scientific discoveries. we introduce the AxonEM dataset, which consists of two 30×30×30 µm^3 EM image volumes from the human and mouse cortex, respectively. We thoroughly proofread over 18,000 axon instances to provide dense 3D axon instance segmentation, enabling largescale evaluation of axon reconstruction methods. In addition, we densely annotate nine ground truth subvolumes for training, per each data volume.
NucMM Dataset: Neuronal Nuclei Segmentation at Sub-Cubic Millimeter Scale
Segmenting 3D cell nuclei from microscopy image volumes is critical for biological and clinical analysis, enabling the study of cellular expression patterns and cell lineages. We pushed the task forward to the sub-cubic millimeter scale and curated the NucMM dataset with two fully annotated volumes: one 0.1 mm^3 electron microscopy (EM) volume containing nearly the entire zebrafish brain with around 170,000 nuclei; and one 0.25 mm^3 micro-CT (uCT) volume containing part of a mouse visual cortex with about 7,000 nuclei.
An Interactive Approach to Train Deep Neural Networks for Segmentation of Neuronal Structures
A web-based interactive tool for training deep neural networks for segmentation tasks. It consists of a server backend that runs on a high-performance GPU compute node, and a front end user interface that runs on a web browser. User configures a classifier, the classes of objects to be detected, and the set of images to use for training and validation.
Source code: https://github.com/Rhoana/icon
Efficient Compression of Segmentation Data For Connectomics
Compresso is a new compression scheme for label data that outperforms existing approaches by using a sliding window to exploit redundancy across border regions in 2D and 3D.
Source code: https://github.com/VCG/compresso
Dense Automatic Neural Annotation
RhoANA is software for dense automatic annotation of neurons in EM serial sections. It includes a processing pipeline, as well as Mojo, a proofreading and annotation tool.
Source code: https://github.com/Rhoana
Distributed Proofreading of Automatic Segmentations
Dojo is a web-based software for proofreading and annotating automatic segmentations of neurons in EM serial sections. It supports collaborative editing of labeled image data in 2D and 3D.
Source code: https://github.com/Rhoana/dojo