Title: Fully automated neuron tracing with deep reinforcement learning
Project Overview: Neuron reconstruction from the whole mouse brain image is a tedious work. During past years, there are many automatic neuron tracing algorithms have been published but those algorithms often fail in reconstruction of the distal arbor and dendrite regions. We have exploited the deep reinforcement learning, which is widely used in self-driving research and applications, to build a self-tracing prototype on the binary data. We expect to extend this study to the real mouse brain data and achieve a relative accurate neuron reconstruction that can be used for further analysis, such as morphology analysis, cell type clustering, etc.
Educational Objectives: Through this task, the candidate will • improve his/her programming skills, especially in C/C++ and python; • learn about basic image processing and analysis techniques; • familiar with big image visualization and annotation infrastructure, Vaa3D-TeraFly; • learn deep learning and apply it on a real study case.
Reference:
Deep Reinforcement Learning for Subpixel Neural Tracking, Tianhong Dai, Proceedings of Machine Learning Research, 2019.