Retinal connectome: mobile game and crowdsourcing algorithms for EyeWire II



 DESCRIPTION (provided by applicant): An online community called EyeWire proved that volunteers can be motivated to reconstruct neural circuits through an activity resembling a 3D coloring book. EyeWire helped discover space-time specificity of the wiring from bipolar cells to starburst amacrine cells, which suggested a surprising new model for direction selectivity in the retina. Motivated by this success, we are preparing to launch EyeWire II, which aims to map the entire retinal connectome, yielding the first complete wiring diagram for any region of the mammalian CNS. This ambitious goal will require innovative advances in virtually every component of EyeWire. The underlying electron microscopic image of the retina will be replaced by a new image with increased size and quality. A new artificial intelligence (AI) will be trained using a new software package for 3D deep learning. While the improved AI is expected to reduce the amount of human effort required to reconstruct a neuron, the number of neurons targeted for reconstruction will also increase dramatically. Overall, the absolute amount of human effort required will increase rather than decrease. Therefore it is critical to improve EyeWire's crowdsourcing to (1) mobilize more human effort and (2) to use human effort more efficiently. This project aims to radically improve both aspects, thereby making the retinal connectome achievable by EyeWire II. In Aim 1, we will create a compelling mobile game with the target of engaging 10x more people than the existing EyeWire community. In Aim 2, we will develop and deploy new crowdsourcing algorithms that extract wisdom from the crowd by weighted voting and optimally assign players to tasks. The Aims will be achieved through collaboration between three organizations. Wired Differently, Inc. (WD) is a new Boston-based nonprofit organization dedicated to "citizen neuroscience" that was recently spun out of MIT. WD currently operates EyeWire in collaboration with the Princeton Neuroscience Institute. The Entertainment Technology Center (ETC) at Carnegie Mellon University will offer a project-based class to its master's students to design and prototype new ideas for the mobile game. Both Aims will produce code and algorithms that will be made publicly available, and could have broad impact on citizen science. As the first crowdsourcing of 3D image analysis, the code produced by Aim 1 could be useful for the many other kinds of 3D images found in biomedical research. The crowdsourcing algorithms of Aim 2 are potentially useful for any citizen science project facing the challenge of obtaining accurate and reliable results from a heterogeneous group of volunteers.