AmygdalaGo-BOLT for boundary-aware segmentation of the human amygdala

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate amygdala segmentation is challenging in pediatric neuroimaging due to its small size and complex morphology.
• Existing segmentation tools struggle with reliability in large-scale population studies.

Purpose of the Study:

• To develop AmygdalaGo-BOLT, a boundary-aware deep learning model for accurate and efficient amygdala segmentation in pediatric brain MRI.
• To improve the scalability of amygdala morphology analysis in population neuroscience.

Main Methods:

• Developed AmygdalaGo-BOLT, a deep learning model integrating multiscale features, spatial priors, and self-attention.
• Trained and validated the model on 1,086 manually labeled pediatric MRI scans.
• Assessed generalizability using independent datasets across different imaging centers and age groups.

Main Results:

• AmygdalaGo-BOLT demonstrated strong agreement with expert manual annotations.
• The model significantly improved accuracy and efficiency compared to existing segmentation tools.
• Achieved robust and scalable amygdala morphology analysis.

Conclusions:

• AmygdalaGo-BOLT offers a reliable solution for segmenting the amygdala in pediatric populations.
• The tool facilitates large-scale neuroimaging studies by overcoming manual tracing limitations.
• Enables more efficient and accurate investigation of brain development and structure.