A high-resolution dataset of mouse brain vasculature for deep learning-based reconstruction

Area of Science:

• Neuroscience
• Biomedical Imaging
• Computational Biology

Background:

• Vascular network reconstruction is vital for understanding brain structure, function, metabolism, and disease.
• Current "segment-first, then reconstruct" methods struggle with variations in vessel size, voids, and capillary complexity, limiting accuracy.

Purpose of the Study:

• To introduce and release a comprehensive annotated dataset of mouse brain vasculature.
• To provide a benchmark for developing and evaluating vascular network segmentation and reconstruction algorithms.

Main Methods:

• Acquired 60 3D image blocks (512x512x512) from four mouse brains using fluorescence micro-optical sectioning tomography (fMOST).
• Annotated diverse vascular morphologies, from large vessels to capillaries, focusing on challenging regions.
• Developed a standardized vascular annotation pipeline and tools.

Main Results:

• A high-quality, annotated 3D dataset of mouse brain vasculature is now available.
• The dataset includes detailed annotations for complex and challenging vascular structures.
• Associated annotation tools and a standardized pipeline are provided.

Conclusions:

• The new dataset and tools will advance research in vascular network analysis.
• It serves as a crucial benchmark for improving the accuracy and robustness of reconstruction algorithms.
• Facilitates development of algorithms for brain structure, function, and disease mechanism studies.