GPU-Accelerated RSF Level Set Evolution for Large-Scale Microvascular Segmentation

TL;DR

This work reformulates the Region-Scalable Fitting (RSF) level set model for three-dimensional, large-scale microvascular segmentation and implements a GPU-accelerated, SIMD/SPMD-capable pipeline. By extending RSF to higher dimensions and introducing volume- and voxel-level parallelism, the authors enable efficient segmentation of gigavoxel vascular networks across multiple imaging modalities. They provide seed-surface initialization, rigorous 3D Monte-Carlo validation against manual annotations, and comprehensive GPU/CPU profiling to quantify speedups and robustness to noise. The approach demonstrates accurate, topologically consistent reconstructions of complex microvascular networks while significantly reducing computation time, illustrating practical applicability to whole-organ vascular modeling. This GPU-enabled RSF framework offers a scalable, high-precision tool for 3D microvascular segmentation in emerging high-resolution microscopy data.

Abstract

Microvascular networks are challenging to model because these structures are currently near the diffraction limit for most advanced three-dimensional imaging modalities, including confocal and light sheet microscopy. This makes semantic segmentation difficult, because individual components of these networks fluctuate within the confines of individual pixels. Level set methods are ideally suited to solve this problem by providing surface and topological constraints on the resulting model, however these active contour techniques are extremely time intensive and impractical for terabyte-scale images. We propose a reformulation and implementation of the region-scalable fitting (RSF) level set model that makes it amenable to three-dimensional evaluation using both single-instruction multiple data (SIMD) and single-program multiple-data (SPMD) parallel processing. This enables evaluation of the level set equation on independent regions of the data set using graphics processing units (GPUs), making large-scale segmentation of high-resolution networks practical and inexpensive. We tested this 3D parallel RSF approach on multiple data sets acquired using state-of-the-art imaging techniques to acquire microvascular data, including micro-CT, light sheet fluorescence microscopy (LSFM) and milling microscopy. To assess the performance and accuracy of the RSF model, we conducted a Monte-Carlo-based validation technique to compare results to other segmentation methods. We also provide a rigorous profiling to show the gains in processing speed leveraging parallel hardware. This study showcases the practical application of the RSF model, emphasizing its utility in the challenging domain of segmenting large-scale high-topology network structures with a particular focus on building microvascular models.

Figures (12)

• Figure 1: Calculation of the region-based intensities $r^+$ and $r^-$. (a, d) The input image is multiplied with the appropriate Heaviside function, and then (b, e) convolved with a Gaussian kernel. (c, f) The final images approximate the local intensity of $\Omega$ outside (top) and inside (bottom) of the contour.
• Figure 2: Linear interpolation is used to merge adjacent level set functions (a and b) $\phi_n$ (c) across overlapping curtain regions (purple/red). The merged $\phi_n$ is then used to generate the final contour (d).
• Figure 3: Outline of the proposed 3D RSF algorithm components implemented on the CPU and GPU. The Evolve component is performed completely using CUDA kernels to calculate $E$, with dependencies listed to minimize redundant computation.
• Figure 4: Initializing the level set from an input image (a). Blob detection using the determinant of the Hessian (b) is used to identify candidate seed points (c) on the vascular centerlines. Fast sweeping is then used to determine the positive distance (d) from seed points to create the $\phi_0$.
• Figure 5: Results of RSF method for 2D synthetic vasculature. The curve evolution process from the initial contour (in the first and third columns) to the final contour (in the second and fourth columns) is shown in every row for the corresponding image. We altered the Signal-to-Noise Ratio (SNR) of the ground truth in certain images through two distinct methods. Firstly, we introduced Gaussian noise, and secondly, we modified the intensity along the x-axis to the y-axis for some images and vice versa along the y-axis to the x-axis for others.