webKnossos is an image annotation tool for large-scale Connectomics. As a web-based tool it enables collaboration of many scientists and facilitates data sharing within the community.
This is the first post in a series of webKnossos-related posts. The next posts will be about webKnossos’ efficient annotation and Machine Learning features.
Large-scale Connectomics reconstructs the network of neurons
The human brain has roughly 86 billion neurons and each neuron has an average of 2 thousand synapses to other neurons. However, the connection network that results from that is largely unknown. Connectomics is a field of Brain Research that is interested in this network. Densely reconstructing the wiring of the neurons, will uncover knowledge about the inner workings of the brain, learning processes and diseases. In order to investigate the neural network, scientists reconstruct neurons and synapses from 3D image data of tissue samples.
Due to the required resolution of a few nanometers, this leads to massive datasets. At a resolution of 4×4×40 nm³, a full fly brain (view in webKnossos), the size of a poppy seed, eats up 50 TB of disk space. A cubic millimeter at the same resolution comes in at 1.5 PB. For reference, a full brain of a mouse, which is still a fairly small animal, is roughly 450 mm³ in size and would amount to 675 PB. The scale of data that the scientists operate on is just mind-boggling. These datasets do not fit on single computers and cannot be analyzed by individual people.
Online Exploration of 3D image datasets
In order to work with datasets of that size, collaborative tools are required. Especially web-based tools lend themselves well to collaboration and data sharing. webKnossos is an open-source tool for exploring and annotating large 3D image datasets. We developed webKnossos in collaboration with the Max Planck Institute for Brain Research and published our results in in Nature Methods. It has been built for the Connectomics community but also is used in other domains.
With webKnossos, users can easily browse through large datasets from wherever they are. The data delivery is optimized for low latency and high bandwidth. With the help of downsampling, it is possible to quickly browse from an overview to the desired location in the dataset at a resolution where most cell components are visible.
Collaboration and publication with webKnossos
Because webKnossos is a web-based tool, sharing a dataset is as easy as sending a link to a colleague. During analysis, this link will be token-secured. And when analysis is done, publication is only a flip of a switch away. Authors can even link to exact locations in the raw data from their paper. When the data is published, other scientists can build upon the work and create annotations directly from webKnossos. At demo.webknossos.org, there is already a selection of datasets (such as this full fly brain, this mouse retina or this Zebrafish neuromast) that will grow over time.
Users don’t need to install anything besides a browser on their computers to use webKnossos. This is especially convenient for crowd-sourcing where a large group of users will collaboratively work on a dataset. It would be a tremendous effort to make sure that everybody has the most up-to-date version of the software installed. webKnossos has been developed with crowd-sourcing in mind and has user, project and task management features to support it.
Boergens, Berning et al., webKnossos: efficient online 3D data annotation for connectomics, 2017
Berning et al., SegEM: Efficient Image Analysis for High-Resolution Connectomics, 2015 (including top image)
Zheng, Lauritzen et al., A Complete Electron Microscopy Volume of the Brain of Adult Drosophila melanogaster, 2018