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

Automatic segmentation enables the rapid, reliable, and reproducible delineation of anatomical structures within the human brain from magnetic resonance imaging (MRI) scans. While large animal species, such as sheep, are extensively used in experimental brain research, limited research exists on automating the segmentation of their brain structures, including the cerebrospinal fluid (CSF) spaces, compared to that of humans. The development of increasingly automatic segmentation methods for animal brain models has significant implications for experimental research in the management of human pathologies such as mild traumatic brain injury (mTBI). Sheep serve as an ideal model for such advancements due to their similarities to humans in terms of their gyrencephalic brain structure, their long lifespan, and docile nature, which enhance their utility in translational neuroscience research.Here, we present an atlas-free semi-automatic segmentation method for the CSF spaces in sheep, capable of segmenting manually skull-stripped structural MRI images in under one minute. We use the nn-UNet deep learning framework based on convolutional neural networks to train segmentation models on data from four adult sheep brains. When comparing our predictions to manually segmented ground truths, we achieved a high Dice overlap score of 94.05% for the CSF space as a whole, and 75.08% for the intricate ventricular CSF space alone.Clinical relevance-Our atlas-free semi-automatic sheep CSF space segmentation method offers substantial utility in experimental brain research worldwide and holds potential for clinical application upon translation, pending validation on larger clinical datasets.