Abstract: A robust and efficient method for the segmentation of 3D structures in CT and MR images is presented. The proposed method is based on 3D ray propagation by mean shift analysis with a smoothness constraint. Specifically, ray propagation is used to guide an evolving surface due to its computational efficiency. In addition, non-parametric analysis and shape priors are incorporated to the proposed technique for robust convergence. Several examples are depicted to illustrate its effectiveness.
Introduction
Segmenting 3D structures in Computed Tomography (CT) and Magnetic Resonance (MR) images is a critical yet challenging task in medical imaging. Clinicians often face difficulties due to high noise levels, partial volume effects—where a single voxel contains multiple tissue types—and the significant computational overhead of processing large 3D datasets.
Traditional segmentation techniques typically fall into two categories: Level Set methods and Explicit Deformable Models. While Level Set methods offer topological flexibility, they are computationally intensive. Conversely, Explicit Deformable Models often require complex re-parameterization to prevent surface mesh tangling. Furthermore, both approaches frequently rely on image gradients to identify boundaries, making them highly susceptible to noise. This paper proposes a "one-click" framework designed for real-time clinical use that is robust to fuzzy data and easy for radiologists to initialize.
Methodology
The core of the proposed framework is Robust Ray Propagation, which combines computational efficiency with noise-resilient boundary detection.
Ray Propagation Engine
Instead of evolving a complex 3D grid, the system utilizes a Lagrangian approach, shooting rays outward from a user-defined seed point. This method is highly efficient as it only tracks the movement of sample points along these rays. To ensure an even distribution, the surface is initialized by subdividing an octahedron, creating a sphere-like starting point with equidistant samples.
Mean Shift Analysis
To locate boundaries, the algorithm employs mean shift—a non-parametric clustering technique—to analyze the intensity profile along each ray. Rather than searching for the steepest gradient, it calculates displacement vectors that point away from discontinuities. This non-parametric approach treats edge detection as a density estimation problem, allowing the algorithm to ignore random noise spikes that typically disrupt gradient-based segmenters.
The Speed Function
The expansion of the surface is governed by a speed function $S = S_0 + \beta S_1$:
- $S_0$ (Data Term): Derived from the mean shift analysis, this term causes rays to move quickly through homogeneous areas and slow down or "snap" back upon hitting a boundary.
- $S_1$ (Smoothness Term): This term utilizes mean curvature to ensure the resulting 3D shape is physically plausible, helping the algorithm bridge regions where image data may be missing or blurry.
Key Results and Performance
The method was primarily demonstrated on 3D Computed Tomography Angiography (CTA) data for aneurysm segmentation and contrast-enhanced CT/MR data for general structures.
- Efficiency: The algorithm achieved 3D segmentations in under 10 seconds on a Pentium III 1GHz processor, enabling real-time interaction for clinical workflows.
- Robustness: Qualitative results indicate successful capture of lumen boundaries in noisy CTA data, even where boundaries are poorly defined.
- User Experience: By allowing a radiologist to click once inside a structure to receive an instant volumetric measurement, the tool represents a significant improvement over manual slicing.
Limitations and Future Work
While effective for "blob-like" pathologies such as aneurysms or tumors, the single-source ray propagation method cannot naturally handle toroidal shapes (structures with holes) or complex branching that wraps back on itself. Future work involves rigorous quantitative validation against clinical ground truth and physical phantoms. Additionally, the authors suggest that incorporating more advanced geometric techniques, such as Mean-Gaussian curvature, could further refine results for highly irregular shapes.
Summary updated