Subvoxel accurate graph search using non-Euclidean graph space.

Area of Science:

• Medical image analysis
• Computational anatomy
• Graph theory applications

Background:

• Graph search offers globally optimal solutions for volumetric medical image analysis.
• Traditional graph methods using Euclidean space limit segmentation precision to voxel resolution and ignore partial volume effects.

Purpose of the Study:

• To develop a generalized graph search method for subvoxel accurate segmentation of volumetric medical images.
• To maintain computational efficiency and global optimality while improving segmentation precision.

Main Methods:

• Generalizing graph representation to non-Euclidean space with non-equidistant node spacing.
• Utilizing a deformation field to adaptively adjust regional node density based on expected cost.
• Applying the method to optical coherence tomography (OCT) and 3-D MRI data.

Main Results:

• Achieved statistically significant improvements in segmentation accuracy on OCT and 3-D MR datasets.
• Demonstrated subvoxel accuracy, surpassing the limitations of traditional voxel-based segmentation.
• Maintained computational efficiency and global optimality comparable to existing methods.

Conclusions:

• The generalized graph search method enables subvoxel accurate medical image segmentation.
• This approach enhances accuracy with existing hardware and preserves accuracy with lower-resolution equipment.
• The method is broadly applicable across imaging modalities and extensible to higher dimensions.