André S. E. Koster

Probabilistic multiscale image segmentation

A method is presented to segment multidimensional images using a multiscale (hyperstack) approach with probabilistic linking. A hyperstack is a voxel-based multiscale data structure whose levels are constructed by convolving the original image with a Gaussian kernel of increasing width. Between voxels at adjacent scale levels, child-parent linkages are established according to a model-directed linkage scheme. In the resulting tree-like data structure, roots are formed to indicate the most plausible locations in scale space where segments in the original image are represented by a single voxel. The final segmentation is obtained by tracing back the linkages for all roots. The present paper deals with probabilistic (or multiparent) linking. The multiparent linkage structure is translated into a list of probabilities that are indicative of which voxels are partial volume voxels and to which extent. Probability maps are generated to visualize the progress of weak linkages in scale space when going from fine to coarser scale. It is demonstrated that probabilistic linking gives a significantly improved segmentation as compared with conventional (single-parent) linking.

Probabilistic segmentation of partial volume voxels

Abstract A multiscale method is proposed to detect and classify partial volume voxels in a multi-dimensional image. The result — a list of probabilities for each partial volume voxel — makes measurements of image geometry more reliable and improves volumetric visualization of image objects.

Heuristic linking models in multiscale image segmentation

This paper presents a novel approach to multiscale image segmentation. It addresses the linking of pixels at adjacent levels in scale-space and the labeling of roots representing segments in the original image. In previous multiscale segmentation approaches, linking and root labeling were based on intensity proximity only. The approach proposed here contains multiple heuristic mechanisms that result in a single criterion for linking (affection) and root labeling (adultness). The segmentations are validated by measuring the amount of postprocessing that is needed to reach an objectively defined accuracy of segmentation. The evaluation is performed using three artificial 2D images with different characteristics, and two 2D magnetic resonance brain images. A comparison is made with a pyramid segmentation method. It is found that several of the proposed heuristic link and root mechanisms improve the performance of multiscale segmentation. A very satisfactory segmentation of all images could be obtained by using a fixed set of compromised weight settings of the most effective mechanisms.

A methodology for the validation of image segmentation methods

A validation methodology for image segmentation methods is explored that is based on quality constrained cost analysis. In this methodology a segmentation method is evaluated by the cost reduction it provides relative to the cost of a full-interactive (manual) segmentation. This cost is constrained by a quality threshold, so that less-than-perfect segmentations are allowed. In this way, segmentation methods which are entirely different in nature can be compared objectively. The validation methodology is presented in its most general form, along with an example of its application to the comparison of segmentation methods.<<ETX>>

Validation of the interleaved pyramid for the segmentation of 3D vector images

Abstract A multiresolution pyramid with double scale space sampling (compared to the Burt & Hong scheme) for the segmentation of 3D images, of which the elements are multiple valued, is described. Evaluation is carried out by quality constrained cost analysis (QCCA).

Probabilistic Hyperstack Segmentation of MR Brain Data

A multiscale method (the hyperstack) is proposed to segment multidimensional MR brain data. Hyperstack segmentation is based upon the linking of voxels at adjacent levels in scale space, followed by a root selection to find the voxels that represent the segments in the original image. This paper addresses an advanced linking and root labeling method for the hyperstack. In particular, attention will be paid to an extension of the linking scheme for the detection and classification of partial volume voxels. The result—a list of probabilities for each partial volume voxel—improves the resulting segmentations.

Multiresolution segmentation of 3D images by the hyperstack

An improvement in the design of the hyperstack, a three-dimensional image segmentation tool, is described. It is based on a multiresolution approach which is mathematically supported by the diffusion equation. The blurring strategy, used to build the scale space, is outlined, including some difficulties that occur in view of the transition from 2-D to 3-D. The existing prototype, a hyperstack grounded on isointensity following, is extended by two novel ideas: weighted linking and stand-alone parents. The result of a segmented 3D SPECT image (of a liver) is shown. Theoretical considerations concerning the addition of feature information to guide the segmentation process are briefly mentioned. A flexible way to obtain several output images from one single hyperstack is outlined and the reduction of the sampling rate by means of interpolation, which will decrease the total amount of processing time, is investigated.<<ETX>>

Task-directed evaluation of image segmentation methods

In the image processing literature many methods to segment 2D and 3D images have been presented. However, relatively little effort has been spent on the validation of the results of these methods. The goal of the paper is to explore a validation methodology that is based on developing a task-directed quality norm that can be used as a constraint in cost analysis. In this methodology a segmentation method is evaluated by the cost reduction it provides relative to the cost of a full-interactive (manual) segmentation. This cost is constrained by a quality threshold, so that less-than-perfect segmentations are allowed. In this way segmentation methods can be compared, which are designed for the same task, but are different of nature.<<ETX>>

Multiscale Image Segmentation based on a Receptive Field Model

A multiscale method (the hyperstack) is proposed to segment multidimensional image data. Hyperstack segmentation is based on the scale space concept of front-end vision. A scale space is created by blurring the original image using linear Gaussian kernels of increasing width.

Probabilistic multiscale image segmentation: set-up and first results (Proceedings Only)

We have developed a method to segment two- and three-dimensional images using a multiscale (hyperstack) approach with probabilistic linking. A hyperstack is a voxel-based multiscale data structure containing linkages between voxels at different scales. The scale-space is constructed by repeatedly applying a discrete convolution with a Gaussian kernel to the original input image. Between these levels of increasing scale we establish child-parent linkages according to a linkage scheme that is based on affection. In the resulting tree-like data structure roots are formed to indicate the most plausible locations in scale-space where objects (of different sizes) are actually defined by a single voxel. Tracing the linkages back from every root to the ground level produces a segmented image. The present paper deals with probabilistic linking, i.e., a set-up in which a child voxel can be linked to more than one parent voxel. The output of the thus constructed hyperstack -- a list of object probabilities per voxel -- can be directly related to the opacities used in volume renderers.