Evaluation of automatic neonatal brain segmentation algorithms: the NeoBrainS12 challenge

Area of Science:

• Medical Imaging
• Neuroimaging
• Computational Neuroscience

Background:

• Accurate brain segmentation is crucial for analyzing neurodevelopment in preterm infants.
• Existing brain segmentation algorithms lack standardized performance comparisons.
• The NeoBrainS12 study addresses this gap by evaluating multiple algorithms on diverse infant brain MRI datasets.

Purpose of the Study:

• To compare the performance of different brain segmentation algorithms in preterm infants.
• To evaluate segmentation accuracy across various brain tissues and MRI scan types.
• To identify limitations of current algorithms, particularly for myelinated white matter.

Main Methods:

• Utilized three distinct MRI datasets from preterm infants (axial and coronal scans at 30 and 40 weeks corrected age).
• Eight participating teams applied their segmentation algorithms to T1- and T2-weighted 3T MRI scans.
• Segmentations included cortical grey matter, white matter (myelinated and non-myelinated), brainstem, basal ganglia, thalami, cerebellum, and cerebrospinal fluid.

Main Results:

• Participating algorithms demonstrated high accuracy in segmenting most brain tissue classes.
• Segmentation of myelinated white matter proved challenging for all evaluated methods.
• Performance varied across different scan types and corrected ages.

Conclusions:

• Current brain segmentation algorithms are effective for most infant brain tissues but require improvement for myelinated white matter.
• The NeoBrainS12 study provides a valuable benchmark for future algorithm development in infant neuroimaging.
• Further research is needed to enhance the segmentation of specific white matter components in developing brains.