Robust atlas-based brain segmentation using multi-structure confidence-weighted registration.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Accurate brain segmentation is crucial for neurological research and clinical diagnosis.
• Existing atlas-based methods often struggle with anatomical variability and registration errors.
• Developing robust segmentation techniques is essential for quantitative analysis of brain structures.

Purpose of the Study:

• To develop and validate a novel, robust, and accurate atlas-based brain segmentation method.
• To improve image registration and achieve anatomically constrained correspondence using multiple initial segmentations.
• To incorporate segmentation confidence maps (SCMs) and supervised atlas correction to enhance accuracy.

Main Methods:

• Utilized multiple initial structure segmentations to simultaneously drive image registration.
• Derived segmentation confidence maps (SCMs) from manual segmentations to weight registration influence.
• Employed a supervised atlas correction technique to address potential registration errors.
• Applied the method to amygdala segmentation in autistic patients and controls, and to eight subcortical structures in MRI datasets.

Main Results:

• Achieved a Dice overlap score of 0.84 for amygdala segmentation in autistic patients and controls.
• Demonstrated results comparable or superior to competing methods for segmenting eight subcortical structures.
• The method effectively incorporated segmentation confidence and atlas correction for improved accuracy.

Conclusions:

• The proposed atlas-based segmentation method is robust and accurate for brain structure segmentation.
• The integration of SCMs and supervised atlas correction enhances segmentation reliability.
• This technique holds significant potential for neuroimaging research and clinical applications, particularly in conditions like autism.