Efficient algorithms for segmenting globally optimal and smooth multi-surfaces.

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Anatomy

Background:

• Accurate segmentation of single and multiple surfaces in 3D volumetric medical images is crucial but remains challenging, particularly with noisy and edge-weak data.
• Existing methods struggle to balance global optimality and surface smoothness effectively.

Purpose of the Study:

• To develop a novel, efficient, graph-theoretical iterative method for globally optimal and smooth segmentation of single and multiple surfaces in 3D volumetric images.
• To incorporate curvature information into the segmentation process to enhance surface smoothness.

Main Methods:

• A graph-theoretical iterative approach using a volumetric graph representation of 3D images.
• Incorporation of curvature information into the objective function to ensure surface smoothness.
• Evaluation on 14 3D Optical Coherence Tomography (OCT) images.

Main Results:

• The proposed method achieves optimal or near-optimal solutions within 3 to 5 iterations.
• It is the first algorithm to use curvature in the objective function for global optimality and smoothness.
• Demonstrates significantly improved running time compared to existing methods.
• Achieves comparable global optimality with substantially enhanced smoothness, ideal for noisy images.

Conclusions:

• The novel graph-theoretical method offers an efficient and effective solution for 3D surface segmentation in medical imaging.
• The integration of curvature ensures superior surface smoothness, making it highly suitable for challenging, noisy datasets.
• This approach advances the state-of-the-art in medical image segmentation by optimizing both accuracy and smoothness.