Medial axis segmentation of cranial nerves using shape statistics-aware discrete deformable models.

Area of Science:

• Medical Imaging
• Neuroscience
• Computational Anatomy

Background:

• Accurate segmentation of brainstem cranial nerves is vital for neurosurgical planning.
• Existing methods may struggle with low-resolution MRI and image artifacts.

Purpose of the Study:

• To develop a robust segmentation methodology for brainstem cranial nerves using statistical shape models (SSM) and deformable 3D contours.
• To create a probabilistic digital atlas of ten cranial nerve pairs (CNIII-CNXII).

Main Methods:

• Utilized a 1-Simplex discrete deformable 3D contour model for segmenting nerve centerlines from T2 MR images.
• Developed shape models for ten cranial nerve pairs, incorporating shape information as prior knowledge.
• Employed an entropy-based energy minimization framework for point correspondence and a shape-based internal force for stability.

Main Results:

• Constructed ten SSMs for brainstem cranial nerves, assessed for compactness, specificity, and generality.
• Achieved sub-voxel accuracy with mean absolute shape distance (MASD) of 0.19 ± 0.13 mm and Hausdorff distance (HD) of 0.21 mm.
• Demonstrated robustness and stability using synthetic and patient MRI data, even with low resolution and artifacts.

Conclusions:

• The proposed methodology integrates SSM with deformable contours for accurate and robust centerline segmentation of cranial nerves.
• This approach is essential for creating precise 3D patient-specific models for neurosurgical planning and simulations.
• The resulting probabilistic digital atlas enhances the understanding and visualization of critical neuroanatomical structures.