Tobias Heimann

Statistical shape models for 3D medical image segmentation: a review.

Statistical shape models (SSMs) have by now been firmly established as a robust tool for segmentation of medical images. While 2D models have been in use since the early 1990 s, wide-spread utilization of three-dimensional models appeared only in recent years, primarily made possible by breakthroughs in automatic detection of shape correspondences. In this article, we review the techniques required to create and employ these 3D SSMs. While we concentrate on landmark-based shape representations and thoroughly examine the most popular variants of Active Shape and Active Appearance models, we also describe several alternative approaches to statistical shape modeling. Structured into the topics of shape representation, model construction, shape correspondence, local appearance models and search algorithms, we present an overview of the current state of the art in the field. We conclude with a survey of applications in the medical field and a discussion of future developments.

A shape-guided deformable model with evolutionary algorithm initialization for 3D soft tissue segmentation

We present a novel method for the segmentation of volumetric images, which is especially suitable for highly variable soft tissue structures. Core of the algorithm is a statistical shape model (SSM) of the structure of interest. A global search with an evolutionary algorithm is employed to detect suitable initial parameters for the model, which are subsequently optimized by a local search similar to the Active Shape mechanism. After that, a deformable mesh with the same topology as the SSM is used for the final segmentation: While external forces strive to maximize the posterior probability of the mesh given the local appearance around the boundary, internal forces governed by tension and rigidity terms keep the shape similar to the underlying SSM. To prevent outliers and increase robustness, we determine the applied external forces by an algorithm for optimal surface detection with smoothness constraints. The approach is evaluated on 54 CT images of the liver and reaches an average surface distance of 1.6 +/- 0.5 mm in comparison to manual reference segmentations.

Active shape models for a fully automated 3d segmentation of the liver – an evaluation on clinical data

This paper presents an evaluation of the performance of a three-dimensional Active Shape Model (ASM) to segment the liver in 48 clinical CT scans. The employed shape model is built from 32 samples using an optimization approach based on the minimum description length (MDL). Three different gray-value appearance models (plain intensity, gradient and normalized gradient profiles) are created to guide the search. The employed segmentation techniques are ASM search with 10 and 30 modes of variation and a deformable model coupled to a shape model with 10 modes of variation. To assess the segmentation performance, the obtained results are compared to manual segmentations with four different measures (overlap, average distance, RMS distance and ratio of deviations larger 5mm). The only appearance model delivering usable results is the normalized gradient profile. The deformable model search achieves the best results, followed by the ASM search with 30 modes. Overall, statistical shape modeling delivers very promising results for a fully automated segmentation of the liver.

Validation of Fourier decomposition MRI with dynamic contrast-enhanced MRI using visual and automated scoring of pulmonary perfusion in young cystic fibrosis patients

PURPOSE To validate Fourier decomposition (FD) magnetic resonance (MR) imaging in cystic fibrosis (CF) patients with dynamic contrast-enhanced (DCE) MR imaging. MATERIALS AND METHODS Thirty-four CF patients (median age 4.08 years; range 0.16-30) were examined on a 1.5-T MR imager. For FD MR imaging, sets of lung images were acquired using an untriggered two-dimensional balanced steady-state free precession sequence. Perfusion-weighted images were obtained after correction of the breathing displacement and Fourier analysis of the cardiac frequency from the time-resolved data sets. DCE data sets were acquired with a three-dimensional gradient echo sequence. The FD and DCE images were visually assessed for perfusion defects by two readers independently (R1, R2) using a field based scoring system (0-12). Software was used for perfusion impairment evaluation (R3) of segmented lung images using an automated threshold. Both imaging and evaluation methods were compared for agreement and tested for concordance between FD and DCE imaging. RESULTS Good or acceptable intra-reader agreement was found between FD and DCE for visual and automated scoring: R1 upper and lower limits of agreement (ULA, LLA): 2.72, -2.5; R2: ULA, LLA: ± 2.5; R3: ULA: 1.5, LLA: -2. A high concordance was found between visual and automated scoring (FD: 70-80%, DCE: 73-84%). CONCLUSIONS FD MR imaging provides equivalent diagnostic information to DCE MR imaging in CF patients. Automated assessment of regional perfusion defects using FD and DCE MR imaging is comparable to visual scoring but allows for percentage-based analysis.

Automatic quantification of subcutaneous and visceral adipose tissue from whole‐body magnetic resonance images suitable for large cohort studies

To develop an automated method with which to distinguish metabolically different adipose tissues in a large number of subjects using whole‐body magnetic resonance imaging (MRI) datasets for improving the understanding of chronic disease risk predictions associated with distinct adipose tissue compartments.

3D active shape models using gradient descent optimization of description length

Active Shape Models are a popular method for segmenting three-dimensional medical images. To obtain the required landmark correspondences, various automatic approaches have been proposed. In this work, we present an improved version of minimizing the description length (MDL) of the model. To initialize the algorithm, we describe a method to distribute landmarks on the training shapes using a conformal parameterization function. Next, we introduce a novel procedure to modify landmark positions locally without disturbing established correspondences. We employ a gradient descent optimization to minimize the MDL cost function, speeding up automatic model building by several orders of magnitude when compared to the original MDL approach. The necessary gradient information is estimated from a singular value decomposition, a more accurate technique to calculate the PCA than the commonly used eigendecomposition of the covariance matrix. Finally, we present results for several synthetic and real-world datasets demonstrating that our procedure generates models of significantly better quality in a fraction of the time needed by previous approaches.

Branching patterns and drainage territories of the middle hepatic vein in computer-simulated right living-donor hepatectomies.

Full right hepatic grafts are most frequently used for adult‐to‐adult living donor liver transplantation (LDLT). One of the major problems is venous drainage of segments 5 and 8. Thus, this study was designed to provide information on venous drainage of right liver lobes for operation‐planning. Fifty‐six CT data sets from routine clinical imaging were evaluated retrospectively using a liver operation‐planning system. We defined and analyzed venous drainage segments and the impact of anatomic variations of the middle hepatic vein (MHV) on venous outflow from segments 5 and 8. MHV variations led to significant shifts of segment 5 drainage between the middle and right hepatic vein. In cases with the most frequent MHV branching pattern (n = 33), a virtual hepatectomy closely right to the MHV intersected drainage vessels that provided drainage for 30% of the potential graft, not taking into account potential veno‐venous shunts. In individuals with inferior MHV branches that extend far into segments 5 and 6 (n = 10), the overall graft volume at risk of impaired venous drainage increased by 5% (p < 0.001). If this is confirmed in clinical trials and correlated with intraoperative findings, the use of liver operation‐planning systems would be beneficial to improve overall outcome after right lobe LDLT.

Lymph node segmentation on CT images by a shape model guided deformable surface method

With many tumor entities, quantitative assessment of lymph node growth over time is important to make therapy choices or to evaluate new therapies. The clinical standard is to document diameters on transversal slices, which is not the best measure for a volume. We present a new algorithm to segment (metastatic) lymph nodes and evaluate the algorithm with 29 lymph nodes in clinical CT images. The algorithm is based on a deformable surface search, which uses statistical shape models to restrict free deformation. To model lymph nodes, we construct an ellipsoid shape model, which strives for a surface with strong gradients and user-defined gray values. The algorithm is integrated into an application, which also allows interactive correction of the segmentation results. The evaluation shows that the algorithm gives good results in the majority of cases and is comparable to time-consuming manual segmentation. The median volume error was 10.1% of the reference volume before and 6.1% after manual correction. Integrated into an application, it is possible to perform lymph node volumetry for a whole patient within the 10 to 15 minutes time limit imposed by clinical routine.

Automatic generation of 3D statistical shape models with optimal landmark distributions.

OBJECTIVES To point out the problem of non-uniform landmark placement in statistical shape modeling, to present an improved method for generating landmarks in the 3D case and to propose an unbiased evaluation metric to determine model quality. METHODS Our approach minimizes a cost function based on the minimum description length (MDL) of the shape model to optimize landmark correspondences over the training set. In addition to the standard technique, we employ an extended remeshing method to change the landmark distribution without losing correspondences, thus ensuring a uniform distribution over all training samples. To break the dependency of the established evaluation measures generalization and specificity from the landmark distribution, we change the internal metric from landmark distance to volumetric overlap. RESULTS Redistributing landmarks to an equally spaced distribution during the model construction phase improves the quality of the resulting models significantly if the shapes feature prominent bulges or other complex geometry. CONCLUSIONS The distribution of landmarks on the training shapes is -- beyond the correspondence issue -- a crucial point in model construction.

Interactive Surface Correction for 3D Shape Based Segmentation

Statistical shape models have become a fast and robust method for segmentation of anatomical structures in medical image volumes. In clinical practice, however, pathological cases and image artifacts can lead to local deviations of the detected contour from the true object boundary. These deviations have to be corrected manually. We present an intuitively applicable solution for surface interaction based on Gaussian deformation kernels. The method is evaluated by two radiological experts on segmentations of the liver in contrast-enhanced CT images and of the left heart ventricle (LV) in MRI data. For both applications, five datasets are segmented automatically using deformable shape models, and the resulting surfaces are corrected manually. The interactive correction step improves the average surface distance against ground truth from 2.43mm to 2.17mm for the liver, and from 2.71mm to 1.34mm for the LV. We expect this method to raise the acceptance of automatic segmentation methods in clinical application.