Template-Based Label Propagation for Mouse Brain MRI Skull Stripping.

Area of Science:

• Biomedical Imaging
• Neuroscience
• Computational Biology

Background:

• Accurate skull stripping is crucial for mouse brain MRI analysis, but current methods are labor-intensive and variable.
• Existing techniques often require manual brain masks for numerous subjects, hindering large-scale studies.

Purpose of the Study:

• To develop a high-throughput, automated skull stripping pipeline for mouse brain MRI.
• To reduce the manual annotation effort required for generating training data for segmentation models.

Main Methods:

• A template-based label propagation approach was used to generate training data.
• An average ex vivo MRI template was created, and a single brain mask was propagated to subjects.
• An attention-based 3D U-Net model was trained using the propagated labels.

Main Results:

• The proposed pipeline achieved competitive segmentation performance while significantly decreasing manual annotation time.
• Training with propagated labels alone demonstrated robust performance, highlighting the importance of label consistency.
• The framework showed adaptability across different imaging conditions (ex vivo to in vivo) when the full pipeline was applied.

Conclusions:

• The developed pipeline offers a practical strategy for efficient, large-scale generation of anatomically consistent training datasets for mouse brain MRI segmentation.
• The template-based label propagation method is effective in reducing manual effort and achieving robust skull stripping.
• The framework's adaptability across imaging domains makes it a valuable tool for diverse mouse brain MRI studies.