Brightfield Auto-Reconstruction Challenge
Simultaneous with the BioImage Informatics 2019 conference, the Allen Institute will hold a challenge for interested labs to contribute to the field of auto-reconstruction of neuron morphology, using our large dataset of brightfield images and manually curated reconstructions.
Cell types are the cellular building blocks of the brain. At the Allen Institute, we are interested in identifying all the cell types in the mouse brain and human cortex. A robust description of cell types must be based on multiple cellular properties. We characterize neurons based on their transcriptomic, electrophysiological and morphological properties. To measure these properties at the single cell level, we optimized an acute slice, patchclamp method (Patchseq) to collect triple modality data at an unprecedented scale of multiple 1000s of cells per year. With this method, transcriptomics and electrophysiology data are relatively quick to collect, but morphology data lags far behind due in part to the labor-intensive process of drawing each neuron manually. Though much work has been done to automate the neuron reconstruction process, most efforts have been focused on fluorescence-based images, which are not compatible with our morphology collection pipeline.
With the Allen Institute pipeline, biocytin-filled neurons are visualized with a HRP-DAB reaction, then imaged in 3D with a brightfield microscope. We use brightfield imaging because it is fast, inexpensive and straight-forward to use. In addition, the DAB reaction has the advantage of high sensitivity and high stability allowing for long-term preservation of labeled tissue.
With this challenge, we hope to encourage the community to use our large dataset of brightfield images and manually curated reconstructions to develop an automated method for representing the morphology of biocytin-filled cells imaged with a brightfield microscope that can keep pace with the scale of our Patchseq pipeline. With the development of such a tool, we have the potential to generate the largest cortical morphology dataset to date, and to use this dataset in part of a comprehensive description of cell types in the mouse and human brain.
FAQ for Participants
- All challenge participants must be registered for the BioImage Informatics 2019 meeting.
- To join the challenge, participants must email the challenge team using the link below, at which time they will be provided access to the training dataset, which includes 3D image stacks and manually generated reconstructions of biocytin-filled neurons, and may begin training their algorithms.
- The deadline to join the challenge is September 30, 2019.
- All participants will be provided access to a separate, competition dataset of 3D image stacks on which to test their algorithm on Wednesday, September 18.
- Participants will be responsible to provide results from the competition dataset as SWC files to the challenge team no later than Wednesday, October 2. All participants are expected to share their implementation publicly.
- Participants must attend the Brightfield Auto-Reconstruction Competition presentation, judging and award session from 12-2pm on Friday, October 4, at the end of the BioImage Informatics meeting.
- Judges for the competition include:
- Staci Sorensen and Hongkui Zeng, Allen Institute for Brain Science
- BioImage Informatics 2019 Planning Committee
- Prizes for the challenge include $500 + NVIDIA gaming GPU for first place (total value of up to $3,500), $250 for second place, and $100 for third place.
- Detailed contest rules are available here.