Thin structure segmentation and visualization in three-dimensional biomedical images: a shape-based approach

Abstract

This paper presents a shape-based approach in extracting thin structures, such as lines and sheets, from three-dimensional (3D) biomedical images. Of particular interest is the capability to recover cellular structures, such as microtubule spindle fibers and plasma membranes, from laser scanning confocal microscopic (LSCM) data. Hessian-based shape methods are reviewed. A synthesized linear structure is used to evaluate the sensitivity of the multiscale filtering approach in extracting closely positioned fibers. We find that the multiscale approach tends to fuse lines together, which makes it unsuitable for visualizing mouse egg spindle fibers. Single-scale Gaussian filters, balanced between sensitivity and noise resistance, are adopted instead. In addition, through an ellipsoidal Gaussian model, the eigenvalues of the Hessian matrix are quantitatively associated with the standard deviations of the Gaussian model. Existing shape filters are simplified and applied to LSCM data. A significant improvement in extracting closely positioned thin lines is demonstrated by the resultant images. Further, the direct association of shape models and eigenvalues makes the processed images more understandable qualitatively and quantitatively.

Patient-Specific Modeling and Analysis of the Mitral Valve Using 3D-TEE

“…The two methods are complementary, as the leaflets may not always be accurately segmented out by the dynamic contour approach, and the heart walls and valve annulus are generally not found by the thin tissue detector. The thin tissue detector models the local TEE intensity as Gaussian and then performs an analysis of the disparities of the eigenvalues associated with the intensity Hessian [25]. If one these eigenvalues is small when compared to the other two, this suggests the presence of a sheet or thin tissue structure.…”

Enhancing the Interactive Visualization of Procedurally Encoded Multifield Data with Ellipsoidal Basis Functions

“…Li et al [LWP * 04] proposed a fitting method using ellipsoids for implicit surfaces with a limited number of data points. Recently Huang et al [HFBC06] showed a shape-based approach for thin structure segmentation and visualization in biomedical images using an ellipsoidal Gaussian model.…”

Virtual Contrast for Coronary Vessels Based on Level Set Generated Subvoxel Accurate Centerlines

“…Given the eigenvalues λ 3 ≤ λ 2 ≤ λ 1 of the 3 × 3 Hessian matrix for each 3D image pixel, it is possible to compute a likelihood of the pixel being part of a linear structure [23,24]. This measure, which we …”

“…This vesselness υ has been used before to improve visualizations of linear structures [23,24], but we are using it to assist in vessel tracking. However, other similar equations have been used for vessel identification before [25].…”