Baby Open Brains: An open-source dataset of infant brain segmentations

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Computational Neuroscience

Background:

• Reproducibility in infant brain development research is hindered by variable processing methods.
• Lack of gold-standard benchmarks, like expert-curated brain segmentations, impedes progress in developing reproducible neuroimaging pipelines.
• Manual infant brain segmentation is complex, demanding significant neuroanatomical expertise and time.

Purpose of the Study:

• To introduce the Baby Open Brains (BOBs) Dataset, a novel open-source resource.
• To provide manually curated and expert-reviewed infant brain segmentations to serve as a benchmark.
• To facilitate the evaluation and enhancement of neuroimaging processing pipelines for infant populations.

Main Methods:

• Collected anatomical MRI data (T1w and T2w) from 71 infant imaging visits across 51 participants (ages 1-9 months).
• Performed manual segmentation of brain structures by experts.
• Ensured expert review of all segmentations for quality control.

Main Results:

• The Baby Open Brains (BOBs) Dataset comprises expert-reviewed segmentations from diverse infant brain MRI scans.
• The dataset captures significant developmental changes in myelination and image intensities from 1 to 9 months of age.
• Established a valuable benchmark for assessing segmentation accuracy and pipeline performance in early life.

Conclusions:

• The BOBs Dataset addresses the critical need for standardized, high-quality infant brain segmentations.
• This resource will accelerate the development of reproducible neuroimaging pipelines for early-life research.
• Provides a foundational dataset for large-scale infant neurodevelopment studies, such as the Human Connectome Project Development (HCP-D).