Robust Automated Amygdala Segmentation via Multi-Atlas Diffeomorphic Registration.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Accurate volumetric segmentation of the amygdala is crucial for understanding brain structure and function.
• Existing segmentation methods can be time-consuming or lack precision for large datasets.

Purpose of the Study:

• To develop and validate a novel, automated method for volumetric segmentation of the human amygdala from MRI data.
• To adapt and improve upon existing open-source segmentation techniques for enhanced amygdala analysis.

Main Methods:

• Utilized multi-atlas segmentation combined with machine learning-based correction.
• Adapted techniques previously applied to hippocampal segmentation.
• Validated the automated method against hand-traced amygdala volumes in 35 human subjects.

Main Results:

• Achieved high Dice coefficients (mean ≈ 0.917) and Jaccard coefficients (mean ≈ 0.848) for automated amygdala segmentation.
• Demonstrated high intra-class correlations (consistency ≈ 0.808, absolute agreement ≈ 0.801) and bivariate correlations (r ≈ 0.814).
• Results show strong agreement with rigorous manual segmentations.

Conclusions:

• The novel automated method provides accurate and reliable volumetric segmentation of the amygdala.
• This technique offers a significant advantage for neuroimaging research involving large sample sizes.
• The approach shows potential for broad application in studies requiring amygdala quantification.