BrainSeg: a generalized framework for comprehensive multimodal brain tissue segmentation, parcellation, and lesion...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Accurate brain segmentation is crucial for quantitative neuroimaging.
• Existing methods struggle with generalization across human lifespan and diverse imaging modalities.
• Comprehensive Brain Segmentation (CBS) involves tissue segmentation, parcellation, and lesion labeling.

Purpose of the Study:

• To introduce BrainSeg, a novel unified framework for Comprehensive Brain Segmentation (CBS).
• To develop an adaptable framework for diverse uni- and multimodal neuroimaging data.
• To ensure generalizability across the entire human lifespan without retraining.

Main Methods:

• Utilized large-scale datasets spanning 14 gestational weeks to 100 years (45,998 scans, 26 datasets).
• Employed a proposed synthesis strategy for data augmentation.
• Developed a unified framework adaptable to diverse input scenarios.

Main Results:

• Achieved state-of-the-art performance in tissue segmentation, brain parcellation, and lesion labeling.
• Internal validation showed averaged Dice ratios up to 96.94% (tissue), 94.25% (parcellation), and 91.06% (lesion).
• External validation demonstrated high accuracy with averaged Dice ratios of 94.01% (tissue) and 91.20% (parcellation).

Conclusions:

• BrainSeg demonstrates robustness and generalizability across diverse neuroimaging conditions.
• The framework provides flexible and reliable analysis for large-scale neuroimaging studies.
• BrainSeg serves as a versatile foundational tool for quantitative neuroimaging.