Skip to main content
Open Access Publications from the University of California

WebStem : Supervision Tool to Improve Unsupervised Landmark Based Registration of Brainstem Sections


Some of the most fundamental tools for research in mouse brain studies are brightfield and fluorescent sections of whole brain sections. The current method of extracting this information, however, requires many hours of human labor. There is a need to automate this process and eliminate or reduce the human interaction aspect. With digital imaging and sectioning becoming more reliable, such a tool can be a "digital brain atlas". Tools, such as the "digital brain atlas", also need to be easily compatible with other tools. Outputting or inputting from one tool or another should require as little human interaction as possible. A brain-wide map annotated with cell types and tract tracing data would allow automatic registration of image stacks to a common coordinate system. Currently, registration of slices requires manual identification of landmarks for every slice. This thesis builds upon the process introduced by "Texture landmark detecting in mouse brain images using significance-based boosting", by Mr. Yuncong Chen. In his paper, he introduced a workflow as a pipeline that analyzed mouse brain textures, segmented them into sections, and classified sections into landmarks using unsupervised or semi-supervised ways. The result of this paper achieves its goal, but there is room for improvement. Since the digital brain atlas is meant to be able to interface with other programs, it needs to obtain high accuracy and reliability. The paper will explain briefly the concepts and methodologies the unsupervised pipeline implements, will introduce a web-app called WebStem to supervise some of the output the pipeline generates, and will show improvements in accuracy thanks to the supervision in three different ways. The new results will serve as the basis for reliable registration and atlas building

Main Content
Current View