Mining neuron morphology is a fundamental task to understand the nerve system and brain working mechanism, since morphology plays a major role in determining neurons’ connectivity and functional properties. Recently, the ever-increasing neuron databases have greatly facilitated the research of neuron morphology. However, the sheer volume and complexity of these data pose significant challenges for computational analysis, preventing the realization of the full potential of these data. The goal of this research project is to seek a new avenue to assemble the massive neuron morphologies and provide a unified framework for neuroscientists to explore and analyze different types of neurons. Particularly, three inter-related components will be implemented for deep and exhaustive neuron mining: 1) accurate and efficient neuron reconstruction and tracing methods based on deep learning models; 2) Efficient discovery of relevant instances among large-sized neuron databases via multi-modal and online binary coding methods; 3) Intelligent and interactive knowledge discovery and mining equipped with latest augmented reality (AR) and mixed reality (MR) techniques. Currently, we developed multiple methods for efficient and accurate neuron retrieval in large-scale, including binary coding, online updating, neuronal feature hierarchy, etc. These methods have validated on more than 58,000 neuron cells, showing promising retrieval performance. Moreover, we demonstrated the use case of our framework in assisting neuroscientists to identify and explore unknown neurons.
@article{li2017indexing, title={Indexing and mining large-scale neuron databases using maximum inner product search}, author={Li, Zhongyu and Fang, Ruogu and Shen, Fumin and Katouzian, Amin and Zhang, Shaoting}, journal={Pattern Recognition}, volume={63}, pages={680--688}, year={2017} } @article{li2017interactive, title={Interactive Exploration for Continuously Expanding Neuron Databases}, author={Li, Zhongyu and Metaxas, Dimitris N and Lu, Aidong and Zhang, Shaoting}, journal={Methods}, volume={115}, pages={100--109}, year={2017} }