Atlas-Free Semi-Automatic Segmentation of Sheep Cerebrospinal Fluid Space from MRI

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Automatic segmentation of brain structures from MRI is crucial for research and clinical applications.
• Limited automated segmentation methods exist for large animal models like sheep, hindering translational neuroscience.
• Sheep brains share structural similarities with human brains, making them valuable models for studying neurological conditions.

Purpose of the Study:

• To develop and validate an atlas-free, semi-automatic segmentation method for cerebrospinal fluid (CSF) spaces in sheep brains using MRI.
• To enable rapid and reproducible analysis of sheep brain anatomy for experimental research.
• To facilitate the translation of findings to human neurological condition research, such as mild traumatic brain injury (mTBI).

Main Methods:

• Utilized the nn-Net deep learning framework, a convolutional neural network-based approach.
• Trained segmentation models on structural MRI data from four adult sheep brains.
• Applied the method to manually skull-stripped MRI images, achieving segmentation in under one minute.

Main Results:

• Achieved a high Dice overlap score of 94.05% for the overall CSF space.
• Obtained a Dice overlap score of 75.08% specifically for the intricate ventricular CSF spaces.
• Demonstrated the method's efficiency with segmentation completion in under one minute per scan.

Conclusions:

• The developed atlas-free semi-automatic method provides a rapid and reliable tool for CSF space segmentation in sheep brains.
• This technique holds significant potential for advancing experimental brain research in translational neuroscience.
• Further validation on larger clinical datasets could pave the way for potential clinical applications.