Groupwise multi-atlas segmentation of the spinal cord's internal structure

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Anatomy

Background:

• Spinal cord internal structure (gray vs. white matter) differentiation is crucial for clinical assessment and prognosis.
• Existing magnetic resonance imaging (MRI) techniques face challenges like low contrast-to-noise ratio and distortions, limiting automated segmentation.
• Inter-subject variability in cervical MRI hinders standard registration and multi-atlas segmentation approaches.

Purpose of the Study:

• To develop a novel, automated, and robust framework for segmenting the internal structure of the cervical spinal cord from MRI data.
• To overcome limitations of existing methods, including poor registration accuracy and lack of automated solutions for spinal cord segmentation.

Main Methods:

• A slice-based groupwise registration framework was developed.
• Key steps include pre-aligning slice-based atlases, constructing a spinal cord variability model, projecting target slices into a low-dimensional space, and estimating segmentation using atlas information.
• The framework incorporates atlas selection and informed parameter initialization.

Main Results:

• The novel framework demonstrated sub-millimetric accuracy in segmenting the cervical spinal cord.
• Significant quantitative and qualitative improvements were observed compared to existing multi-atlas segmentation frameworks.
• The study provided insights into the sensitivity of model parameters.

Conclusions:

• The proposed slice-based groupwise registration framework offers a robust and accurate solution for automated cervical spinal cord MRI segmentation.
• This method addresses critical limitations in current neuroimaging analysis, paving the way for improved clinical assessment and prognosis.
• The framework's performance suggests potential for wider application in spinal cord research and clinical practice.