Consistent cortical reconstruction and multi-atlas brain segmentation.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Whole brain segmentation and cortical surface reconstruction are crucial for human brain investigation.
• Independent execution of these tasks can lead to spatial inconsistencies, hindering integrated analyses.
• Existing methods like FreeSurfer have limitations in specific populations, such as older adults with enlarged ventricles.

Purpose of the Study:

• To introduce a novel method, Multi-atlas CRUISE (MaCRUISE), for self-consistent whole brain segmentation and cortical surface reconstruction.
• To develop a modified version, MaCRUISE(+), for improved performance in the presence of white matter lesions.
• To evaluate the accuracy and robustness of MaCRUISE and MaCRUISE(+) against established benchmarks.

Main Methods:

• Developed MaCRUISE, integrating multi-atlas segmentation with the CRUISE cortical reconstruction method.
• Created MaCRUISE(+) to address challenges posed by white matter lesions.
• Validated surface accuracy using expert-placed cortical landmarks on two independent datasets.
• Assessed segmentation performance on a third dataset with expert volumetric labels.
• Evaluated robustness on 200 MR images from an older adult cohort.

Main Results:

• MaCRUISE constructs self-consistent voxelwise segmentations and cortical surfaces.
• MaCRUISE(+) demonstrates robustness in the presence of white matter pathology.
• MaCRUISE yields more accurate whole brain segmentations compared to independent multi-atlas segmentation.
• MaCRUISE shows comparable accuracy to FreeSurfer, with enhanced robustness in older adults.

Conclusions:

• MaCRUISE provides self-consistent brain segmentations and surfaces, with MaCRUISE(+) offering robustness to white matter pathology.
• The method improves segmentation accuracy and demonstrates superior robustness in older populations compared to FreeSurfer.
• MaCRUISE represents a significant advancement for integrated neuroimaging analyses, particularly in diverse or challenging cohorts.