Anatomical structures segmentation by spherical 3D ray casting and gradient domain editing.

Area of Science:

• Medical Imaging
• Image Segmentation
• Computational Anatomy

Background:

• Accurate segmentation of anatomical structures in medical images is crucial for diagnosis and treatment planning.
• Fuzzy or ill-defined boundaries present a significant challenge for existing segmentation algorithms.
• Current methods often rely on global shape priors or curvature constraints, limiting their applicability.

Purpose of the Study:

• To develop and evaluate a novel image segmentation method capable of handling fuzzy boundaries in medical images.
• To address limitations of existing methods by removing the need for global shape priors or curvature constraints.
• To provide an automatic and robust segmentation technique applicable to various anatomical structures.

Main Methods:

• The proposed method casts 3D rays from a seed point, mapping their lengths to a unit sphere.
• Fuzzy boundary locations are estimated by thresholding the gradient magnitude of these ray lengths.
• True boundaries are derived using Laplacian interpolation on the sphere, offering an automatic stopping criterion.

Main Results:

• Experimental evaluation on 23 kidney segmentations and 16 abdominal aortic aneurysm (AAA) segmentations from CT scans.
• Achieved an average volume overlap error of 12.6% compared to ground-truth segmentations.
• Demonstrated robustness to anatomical variability, noise, and parameter settings.

Conclusions:

• The novel segmentation method effectively addresses the challenge of fuzzy boundaries in medical imaging.
• It offers advantages such as no requirement for global shape priors or curvature constraints and automatic stopping criteria.
• The method provides results comparable to existing techniques while being more robust and versatile.