A Marching Cube Algorithm Based on Edge Growth

Marching Cube algorithm is currently one of the most popular 3D reconstruction surface rendering algorithms. It forms cube voxels through the input image, and then uses 15 basic topological configurations to extract the iso-surfaces in the voxels. It processes each cube voxel in a traversal manner, but it does not consider the relationship between iso-surfaces in adjacent cubes. Due to ambiguity, the final reconstructed model may have holes. We propose a Marching Cube algorithm based on edge growth. The algorithm first extracts seed triangles, then grows the seed triangles and reconstructs the entire 3D model. According to the position of the growth edge, we propose 17 topological configurations with iso-surfaces. From the reconstruction results, the algorithm can reconstruct the 3D model well. When only the main contour of the 3D model needs to be organized, the algorithm performs well. In addition, when there are multiple scattered parts in the data, the algorithm can extract only the 3D contours of the parts connected to the seed by setting the region selected by the seed.

💡 Research Summary

The paper addresses two well‑known drawbacks of the classic Marching Cubes (MC) algorithm—topological ambiguity that leads to holes in the reconstructed surface, and the wasteful processing of the vast majority of empty voxels. To overcome these issues, the authors propose an “edge‑growth” based MC variant that builds the iso‑surface by propagating from a set of seed triangles rather than scanning every cube independently.

The method starts by selecting seed triangles, which are taken from the simplest MC topology (configuration 1, a single triangle). All triangles belonging to one layer of configuration 1 are used as seeds in the experiments. Each edge of a seed triangle becomes a “growth edge” and is placed into a FIFO queue together with its geometric coordinates, gradient values, and the index of the cube to which it belongs. The algorithm then repeatedly dequeues a growth edge, determines the cube that contains it, and decides which of the 17 newly defined topological cases applies. These cases are split into two families: triangle‑edge cases (11 configurations) where the growth edge lies on adjacent cube edges, and quadrilateral‑edge cases (6 configurations) where the edge spans opposite sides of the cube.

A key contribution is the “three‑segment interpolation” used to locate the intersection of the iso‑surface with a cube edge. Given the scalar values at the two edge vertices and the iso‑value, a linear interpolation ratio α is computed. α is then clamped to one of three intervals (α < α₁, α₁ ≤ α ≤ α₂, α > α₂) and replaced by a fixed low, middle (0.5), or high value (αₗ, 0.5, αᵤ). This scheme reduces sensitivity to noise and improves the stability of the generated vertices.

After the intersection points are computed, they are connected to form a triangle (or two triangles for quadrilateral cases). Any newly created edges that have not yet been processed are enqueued as growth edges, and the process continues until the queue is empty. Two label arrays keep track of which cubes have already been visited and which have already contributed growth edges, preventing duplicate work.

The authors evaluate the algorithm on three medical imaging datasets. Compared with the original MC, the edge‑growth method reproduces the main outer contour with far fewer triangles (hundreds versus thousands) and eliminates many of the spurious isolated patches that appear in standard MC results. Moreover, when the data contain multiple disconnected components, only the component connected to the initial seed(s) is reconstructed, demonstrating a built‑in region‑of‑interest capability.

Despite these promising results, several limitations are evident. The selection of seed triangles is currently heuristic; the paper simply uses all configuration‑1 triangles from a single layer, which may not be optimal for noisy or highly complex geometries. No automatic seed‑selection strategy is provided, so user intervention may be required in practice. The 17 topological configurations are illustrated but not formally proven to be exhaustive, leaving uncertainty about handling of more intricate topologies such as tunnels or thin shells. Performance metrics are limited to visual comparisons; quantitative measurements of runtime, memory consumption, and triangle count versus other acceleration techniques (e.g., octree culling, GPU parallelization) are absent. Finally, the parameters of the three‑segment interpolation (α₁, α₂, αₗ, αᵤ) are fixed empirically (0.25, 0.3, 0.7, 0.75) without a sensitivity analysis, which may hinder generalization to datasets with different intensity distributions.

In summary, the paper introduces a novel edge‑growth framework that directly tackles MC’s ambiguity problem by enforcing connectivity through shared edges and by reducing unnecessary voxel processing. The approach is especially attractive for medical imaging scenarios where the primary interest lies in the global shape rather than fine interior detail. Future work should focus on (1) robust, possibly data‑driven seed selection, (2) formal validation of the completeness of the 17‑case topology set, (3) integration with parallel hardware to achieve real‑time performance, and (4) systematic evaluation against existing MC extensions and acceleration schemes.