BIBSNet: A deep learning baby image brain segmentation network for MRI scans

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Medical Image Analysis

Background:

• Infant brain segmentation is crucial for understanding neurodevelopment.
• The rapidly changing infant brain presents unique segmentation challenges for existing algorithms.

Purpose of the Study:

• Introduce BIBSNet (Baby and Infant Brain Segmentation Neural Network), a novel deep neural network for infant brain MRI segmentation.
• Develop a robust and generalizable model for segmenting brain tissues in early postnatal development.

Main Methods:

• Trained BIBSNet on MR brain images from 90 infants (0-8 months) using manually annotated and synthetic data.
• Employed a 10-fold cross-validation procedure for model training and assessment.
• Evaluated performance using Dice Similarity Coefficient (DSC) and compared with Joint Label Fusion (JLF) and iBeat.

Main Results:

• BIBSNet demonstrated superior segmentation accuracy compared to JLF, particularly for gray matter (0.849 vs. 0.713) and white matter (0.862 vs. 0.791).
• BIBSNet-derived anatomical and functional metrics outperformed JLF across most measures.
• No significant difference between BIBSNet and iBeat for infants 0-5 months; iBeat performed better for 6-8 months.

Conclusions:

• BIBSNet offers a significant improvement over JLF for infant brain segmentation across all analyzed age groups.
• BIBSNet is 600x faster than JLF, produces compatible labels, and integrates easily into processing pipelines.
• BIBSNet presents a viable and efficient solution for segmenting infant brains during critical developmental stages.