Fast segmentation with the NextBrain histological atlas.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Structural brain analysis at the subregion level is crucial for understanding healthy aging and neurodegenerative diseases.
• The NextBrain histological atlas aids fine-grained investigations but its segmentation framework is computationally intensive for large studies.

Purpose of the Study:

• To develop and validate an open-source tool that dramatically accelerates brain subregion segmentation.
• To enable high-resolution anatomical analysis at an unprecedented scale and flexibility for large neuroimaging studies.

Main Methods:

• A hybrid approach combining machine learning, contrast-adaptive segmentation, target-specific image synthesis, and fast diffeomorphic registration, all with GPU support.
• The tool enables granular segmentation of brain MRI scans (in vivo or ex vivo) of any resolution and contrast.

Main Results:

• The accelerated approach achieves comparable accuracy to the original method (measured by Dice scores).
• Runtime is reduced by over an order of magnitude, with segmentation completed in under 5 minutes on a GPU.
• Validation was performed on approximately 4,000 brain scans across four modalities (in vivo MRI, ex vivo MRI, HiP-CT, photography).

Conclusions:

• The new tool provides a practical, computationally efficient solution for high-resolution brain subregion segmentation.
• This advancement facilitates large-scale neuroimaging studies, enabling deeper insights into brain structure, aging, and disease.