Filter
Associated Lab
- Remove Kainmueller Lab filter Kainmueller Lab
Publication Date
Type of Publication
19 Publications
Showing 11-19 of 19 resultsDeformable surface models are often represented as triangular meshes in image segmentation applications. For a fast and easily regularized deformation onto the target object boundary, the vertices of the mesh are commonly moved along line segments (typically surface normals). However, in case of high mesh curvature, these lines may intersect with the target boundary at "non-corresponding" positions, or even not at all. Consequently, certain deformations cannot be achieved. We propose an approach that allows each vertex to move not only along a line segment, but within a surrounding sphere. We achieve globally regularized deformations via Markov Random Field optimization. We demonstrate the potential of our approach with experiments on synthetic data, as well as an evaluation on 2 x 106 coronoid processes of the mandible in Cone-Beam CTs, and 56 coccyxes (tailbones) in low-resolution CTs.
In this paper, we propose a mapping from the Auto-context model to a deep Convolutional Neural Network (ConvNet), bridging the gap be- tween these two models, and helping address the challenge of training ConvNets with limited training data.
We present a method for fully automatic segmentation of the bones and cartilages of the human knee from MRI data. Based on statistical shape models and graph-based optimization, first the femoral and tibial bone surfaces are reconstructed. Starting from the bone sur- faces the cartilages are segmented simultaneously with a multi object technique using prior knowledge on the variation of cartilage thickness. We validate our method on 40 clinical MRI datasets acquired before knee replacement.
The Honeybee Brain Atlas serves as 3D database and communicative platform to accumulate structural data, i.e. reconstructed neurons, derived from confocal scans (Brandt et al., 2005) (www.neurobiologie.fu-berlin.de/beebrain/) (1). Transforming neurons into the atlas requires manual segmentation of neuropils within confocal images, a time-consuming task requiring expertise in identifying biological structures which can result in different outcomes from various segmenters.
We present a fully automatic method for 3D segmentation of the mandibular bone from CT data. The method includes an adaptation of statistical shape models of the mandible, the skull base and the midfacial bones, followed by a simultaneous graph-based optimization of adjacent deformable models. The adaptation of the models to the image data is performed according to a heuristic model of the typical intensity distribution in the vincinity of the bone boundary, with special focus on an accurate discrimination of adjacent bones in joint regions. An evaluation of our method based on 18 CT scans shows that a manual correction of the automatic segmentations is not necessary in approx. 60% of the axial slices that contain the mandible.
For biomechanical simulations, the segmentation of multiple adjacent anatomical struc- tures from medical image data is often required. If adjacent structures are barely dis- tinguishable in image data, in general automatic segmentation methods for single struc- tures do not yield sufficiently accurate results. To improve segmentation accuracy in these cases, knowledge about adjacent structures must be exploited. Optimal graph searching (graph cuts) based on deformable surface models allows for a simultaneous segmentation of multiple adjacent objects. However, this method requires a correspon- dence relation between vertices of adjacent surface meshes. Line segments, each con- taining two corresponding vertices, may then serve as shared displacement directions in the segmentation process. In this paper we propose a scheme for constructing a corre- spondence relation in adjacent regions of two arbitrary surfaces. This correspondence relation implies shared displacement directions that we apply for segmentation with de- formable surfaces. Here, overlap of the surfaces is guaranteed not to occur. We show correspondence relations for regions on a femoral head and acetabulum and other adja- cent structures, as well as an evaluation of segmentation results on 50 ct images of the hip joint.
Deformable surface models are often represented as triangular meshes in image segmentation applications. For a fast and easily regularized deformation onto the target object boundary, the vertices of the mesh are commonly moved along line segments (typically surface normals). However, in case of high mesh curvature, these lines may not intersect with the target boundary at all. Consequently, certain deformations cannot be achieved. We propose omnidirectional displacements for deformable surfaces (ODDS) to overcome this limitation. ODDS allow each vertex to move not only along a line segment but within the volumetric inside of a surrounding sphere, and achieve globally optimal deformations subject to local regularization constraints. However, allowing a ball-shaped instead of a linear range of motion per vertex significantly increases runtime and memory. To alleviate this drawback, we propose a hybrid approach, fastODDS, with improved runtime and reduced memory requirements. Furthermore, fastODDS can also cope with simultaneous segmentation of multiple objects. We show the theoretical benefits of ODDS with experiments on synthetic data, and evaluate ODDS and fastODDS quantitatively on clinical image data of the mandible and the hip bones. There, we assess both the global segmentation accuracy as well as local accuracy in high curvature regions, such as the tip-shaped mandibular coronoid processes and the ridge-shaped acetabular rims of the hip bones.
We present a fully automatic 3D segmentation method for the liver from contrast-enhanced CT data. It is based on a combination of a constrained free-form and statistical deformable model. The adap- tation of the model to the image data is performed according to a simple model of the typical intensity distribution around the liver boundary and neighboring anatomical structures, considering the potential presence of tumors in the liver. All parameters of the deformation as well as the initial positioning of the model in the data are estimated automatically.
Object tracking is essential for a multitude of biomedical re- search projects. Automated methods are desired in order to avoid im- possible amounts of manual tracking efforts. However, automatically found solutions are not free of errors, and these errors again have to be identified and resolved manually. We propose six innovative ways for semi-automatic curation of automatically found tracking solutions. Respective user interactions are six simple operations: Inclusion and ex- clusion of objects and tracking decisions, specification of the number of objects, and one-click altering of object segmentations. We show how all proposed interactions can be elegantly incorporated into “assignment models” [1,2,3,4,5,6], an innovative and increasingly popular tracking paradigm. Given some user interaction, the tracking engine is capable of computing the respective globally optimal tracking solution efficiently, even benefitting from “warm start”-capabilities. We show that after in- teractively pointing at a single mistake, multiple segmentation and track- ing errors can be fixed automatically in one single re-evaluation, provably leading to the new, feedback-conscious global optimum.