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

Virtual Landmarks

Much has been published on finding landmarks on object surfaces in the context of shape modeling. While this is still an open problem, many of the challenges of past approaches can be overcome by removing the restriction that landmarks must be on the object surface. The virtual landmarks we propose may reside inside, on the boundary of, or outside the object and are tethered to the object. Our solution is straightforward, simple, and recursive in nature, proceeding from global features initially to local features in later levels to detect landmarks. Principal component analysis (PCA) is used as an engine to recursively subdivide the object region. The object itself may be represented in binary or fuzzy form or with gray values. The method is illustrated in 3D space (although it generalizes readily to spaces of any dimensionality) on four objects (liver, trachea and bronchi, and outer boundaries of left and right lungs along pleura) derived from 5 patient computed tomography (CT) image data sets of the thorax and abdomen. The virtual landmark identification approach seems to work well on different structures in different subjects and seems to detect landmarks that are homologously located in different samples of the same object. The approach guarantees that virtual landmarks are invariant to translation, scaling, and rotation of the object/image. Landmarking techniques are fundamental for many computer vision and image processing applications, and we are currently exploring the use virtual landmarks in automatic anatomy recognition and object analytics.

Comparison of T1-weighted 2D TSE, 3D SPGR, and two-point 3D Dixon MRI for automated segmentation of visceral adipose tissue at 3 Tesla

OBJECTIVES

To evaluate and compare conventional T1-weighted 2D turbo spin echo (TSE), T1-weighted 3D volumetric interpolated breath-hold examination (VIBE), and two-point 3D Dixon-VIBE sequences for automatic segmentation of visceral adipose tissue (VAT) volume at 3 Tesla by measuring and compensating for errors arising from intensity nonuniformity (INU) and partial volume effects (PVE).

MATERIALS AND METHODS

The body trunks of 28 volunteers with body mass index values ranging from 18 to 41.2 kg/m² (30.02 ± 6.63 kg/m²) were scanned at 3 Tesla using three imaging techniques. Automatic methods were applied to reduce INU and PVE and to segment VAT. The automatically segmented VAT volumes obtained from all acquisitions were then statistically and objectively evaluated against the manually segmented (reference) VAT volumes.

RESULTS

Comparing the reference volumes with the VAT volumes automatically segmented over the uncorrected images showed that INU led to an average relative volume difference of -59.22 ± 11.59, 2.21 ± 47.04, and -43.05 ± 5.01 % for the TSE, VIBE, and Dixon images, respectively, while PVE led to average differences of -34.85 ± 19.85, -15.13 ± 11.04, and -33.79 ± 20.38 %. After signal correction, differences of -2.72 ± 6.60, 34.02 ± 36.99, and -2.23 ± 7.58 % were obtained between the reference and the automatically segmented volumes. A paired-sample two-tailed t test revealed no significant difference between the reference and automatically segmented VAT volumes of the corrected TSE (p = 0.614) and Dixon (p = 0.969) images, but showed a significant VAT overestimation using the corrected VIBE images.

CONCLUSION

Under similar imaging conditions and spatial resolution, automatically segmented VAT volumes obtained from the corrected TSE and Dixon images agreed with each other and with the reference volumes. These results demonstrate the efficacy of the signal correction methods and the similar accuracy of TSE and Dixon imaging for automatic volumetry of VAT at 3 Tesla.

Automatic liver segmentation on Computed Tomography using random walkers for treatment planning

Segmentation of the liver from Computed Tomography (CT) volumes plays an important role during the choice of treatment strategies for liver diseases. Despite lots of attention, liver segmentation remains a challenging task due to the lack of visible edges on most boundaries of the liver coupled with high variability of both intensity patterns and anatomical appearances with all these difficulties becoming more prominent in pathological livers. To achieve a more accurate segmentation, a random walker based framework is proposed that can segment contrast-enhanced livers CT images with great accuracy and speed. Based on the location of the right lung lobe, the liver dome is automatically detected thus eliminating the need for manual initialization. The computational requirements are further minimized utilizing rib-caged area segmentation, the liver is then extracted by utilizing random walker method. The proposed method was able to achieve one of the highest accuracies reported in the literature against a mixed healthy and pathological liver dataset compared to other segmentation methods with an overlap error of 4.47 % and dice similarity coefficient of 0.94 while it showed exceptional accuracy on segmenting the pathological livers with an overlap error of 5.95 % and dice similarity coefficient of 0.91.

Entropy-based particle correspondence for shape populations

PURPOSE

Statistical shape analysis of anatomical structures plays an important role in many medical image analysis applications such as understanding the structural changes in anatomy in various stages of growth or disease. Establishing accurate correspondence across object populations is essential for such statistical shape analysis studies.

METHODS

In this paper, we present an entropy-based correspondence framework for computing point-based correspondence among populations of surfaces in a groupwise manner. This robust framework is parameterization-free and computationally efficient. We review the core principles of this method as well as various extensions to deal effectively with surfaces of complex geometry and application-driven correspondence metrics.

RESULTS

We apply our method to synthetic and biological datasets to illustrate the concepts proposed and compare the performance of our framework to existing techniques.

CONCLUSIONS

Through the numerous extensions and variations presented here, we create a very flexible framework that can effectively handle objects of various topologies, multi-object complexes, open surfaces, and objects of complex geometry such as high-curvature regions or extremely thin features.

Particle system based adaptive sampling on spherical parameter space to improve the MDL method for construction of statistical shape models

Minimum description length (MDL) based group-wise registration was a state-of-the-art method to determine the corresponding points of 3D shapes for the construction of statistical shape models (SSMs). However, it suffered from the problem that determined corresponding points did not uniformly spread on original shapes, since corresponding points were obtained by uniformly sampling the aligned shape on the parameterized space of unit sphere. We proposed a particle-system based method to obtain adaptive sampling positions on the unit sphere to resolve this problem. Here, a set of particles was placed on the unit sphere to construct a particle system whose energy was related to the distortions of parameterized meshes. By minimizing this energy, each particle was moved on the unit sphere. When the system became steady, particles were treated as vertices to build a spherical mesh, which was then relaxed to slightly adjust vertices to obtain optimal sampling-positions. We used 47 cases of (left and right) lungs and 50 cases of livers, (left and right) kidneys, and spleens for evaluations. Experiments showed that the proposed method was able to resolve the problem of the original MDL method, and the proposed method performed better in the generalization and specificity tests.

Optimal initialization for 3D correspondence optimization: an evaluation study

The identification of corresponding landmarks across a set of training shapes is a prerequisite for statistical shape model (SSM) construction. We automatically establish 3D correspondence using one new and several known alternative approaches for consistent, shape-preserving, spherical parameterization. The initial correspondence determined by all employed methods is refined by optimizing a groupwise objective function. The quality of all models before and after optimization is thoroughly evaluated using several data sets of clinically relevant, anatomical objects of varying complexity. Correspondence quality is benchmarked in terms of the SSMs' specificity and generalization ability, which are measured using different surface based distance functions. We find that our new approach performs best for complex objects. Furthermore, all new and previously published methods of our own allow for (i) building SSMs that are significantly better than the well-known SPHARM method, (ii) establishing quasi-optimal correspondence for low and moderately complex objects without additional optimization, and (iii) considerably speeding up convergence, thus, providing means for practical, fast, and accurate SSM construction.

Segmenting CT prostate images using population and patient-specific statistics for radiotherapy

PURPOSE

In the segmentation of sequential treatment-time CT prostate images acquired in image-guided radiotherapy, accurately capturing the intrapatient variation of the patient under therapy is more important than capturing interpatient variation. However, using the traditional deformable-model-based segmentation methods, it is difficult to capture intrapatient variation when the number of samples from the same patient is limited. This article presents a new deformable model, designed specifically for segmenting sequential CT images of the prostate, which leverages both population and patient-specific statistics to accurately capture the intrapatient variation of the patient under therapy.

METHODS

The novelty of the proposed method is twofold: First, a weighted combination of gradient and probability distribution function (PDF) features is used to build the appearance model to guide model deformation. The strengths of each feature type are emphasized by dynamically adjusting the weight between the profile-based gradient features and the local-region-based PDF features during the optimization process. An additional novel aspect of the gradient-based features is that, to alleviate the effect of feature inconsistency in the regions of gas and bone adjacent to the prostate, the optimal profile length at each landmark is calculated by statistically investigating the intensity profile in the training set. The resulting gradient-PDF combined feature produces more accurate and robust segmentations than general gradient features. Second, an online learning mechanism is used to build shape and appearance statistics for accurately capturing intrapatient variation.

RESULTS

The performance of the proposed method was evaluated on 306 images of the 24 patients. Compared to traditional gradient features, the proposed gradient-PDF combination features brought 5.2% increment in the success ratio of segmentation (from 94.1% to 99.3%). To evaluate the effectiveness of online learning mechanism, the authors carried out a comparison between partial online update strategy and full online update strategy. Using the full online update strategy, the mean DSC was improved from 86.6% to 89.3% with 2.8% gain. On the basis of full online update strategy, the manual modification before online update strategy was introduced and tested, the best performance was obtained; here, the mean DSC and the mean ASD achieved 92.4% and 1.47 mm, respectively.

CONCLUSIONS

The proposed prostate segmentation method provided accurate and robust segmentation results for CT images even under the situation where the samples of patient under radiotherapy were limited. A conclusion that the proposed method is suitable for clinical application can be drawn.

Building 3-D statistical shape models by direct optimization

Statistical shape models are powerful tools for image interpretation and shape analysis. A simple, yet effective, way of building such models is to capture the statistics of sampled point coordinates over a training set of example shapes. However, a major drawback of this approach is the need to establish a correspondence across the training set. In 2-D, a correspondence is often defined using a set of manually placed 'landmarks' and linear interpolation to sample the shape in between. Such annotation is, however, time-consuming and subjective, particularly when extended to 3-D. In this paper, we show that it is possible to establish a dense correspondence across the whole training set automatically by treating correspondence as an optimization problem. The objective function we use for the optimization is based on the minimum description length principle, which we argue is a criterion that leads to models with good compactness, specificity, and generalization ability. We manipulate correspondence by reparameterizing each training shape. We describe an explicit representation of reparameterization for surfaces in 3-D that makes it impossible to generate an illegal (i.e., not one-to-one) correspondence. We also describe several large-scale optimization strategies for model building, and perform a detailed analysis of each approach. Finally, we derive quantitative measures of model quality, allowing meaningful comparison between models built using different methods. Results are given for several different training sets of 3-D shapes, which show that the minimum description length models perform significantly better than other approaches.

Estimation of pulmonary motion in healthy subjects and patients with intrathoracic tumors using 3D-dynamic MRI: initial results

Objective

To estimate a new technique for quantifying regional lung motion using 3D-MRI in healthy volunteers and to apply the technique in patients with intra- or extrapulmonary tumors.

Materials and Methods

Intraparenchymal lung motion during a whole breathing cycle was quantified in 30 healthy volunteers using 3D-dynamic MRI (FLASH [fast low angle shot] 3D, TRICKS [time-resolved interpolated contrast kinetics]). Qualitative and quantitative vector color maps and cumulative histograms were performed using an introduced semiautomatic algorithm. An analysis of lung motion was performed and correlated with an established 2D-MRI technique for verification. As a proof of concept, the technique was applied in five patients with non-small cell lung cancer (NSCLC) and 5 patients with malignant pleural mesothelioma (MPM).

Results

The correlation between intraparenchymal lung motion of the basal lung parts and the 2D-MRI technique was significant (r = 0.89, p < 0.05). Also, the vector color maps quantitatively illustrated regional lung motion in all healthy volunteers. No differences were observed between both hemithoraces, which was verified by cumulative histograms. The patients with NSCLC showed a local lack of lung motion in the area of the tumor. In the patients with MPM, there was global diminished motion of the tumor bearing hemithorax, which improved siginificantly after chemotherapy (CHT) (assessed by the 2D- and 3D-techniques) (p < 0.01). Using global spirometry, an improvement could also be shown (vital capacity 2.9 ± 0.5 versus 3.4 L ± 0.6, FEV1 0.9 ± 0.2 versus 1.4 ± 0.2 L) after CHT, but this improvement was not significant.

Conclusion

A 3D-dynamic MRI is able to quantify intraparenchymal lung motion. Local and global parenchymal pathologies can be precisely located and might be a new tool used to quantify even slight changes in lung motion (e.g. in therapy monitoring, follow-up studies or even benign lung diseases).

Comparison and evaluation of methods for liver segmentation from CT datasets

This paper presents a comparison study between 10 automatic and six interactive methods for liver segmentation from contrast-enhanced CT images. It is based on results from the "MICCAI 2007 Grand Challenge" workshop, where 16 teams evaluated their algorithms on a common database. A collection of 20 clinical images with reference segmentations was provided to train and tune algorithms in advance. Participants were also allowed to use additional proprietary training data for that purpose. All teams then had to apply their methods to 10 test datasets and submit the obtained results. Employed algorithms include statistical shape models, atlas registration, level-sets, graph-cuts and rule-based systems. All results were compared to reference segmentations five error measures that highlight different aspects of segmentation accuracy. All measures were combined according to a specific scoring system relating the obtained values to human expert variability. In general, interactive methods reached higher average scores than automatic approaches and featured a better consistency of segmentation quality. However, the best automatic methods (mainly based on statistical shape models with some additional free deformation) could compete well on the majority of test images. The study provides an insight in performance of different segmentation approaches under real-world conditions and highlights achievements and limitations of current image analysis techniques.