A method and software for segmentation of anatomic object ensembles by deformable m-reps

Fully automated organ segmentation in male pelvic CT images

Accurate segmentation of prostate and surrounding organs at risk is important for prostate cancer radiotherapy treatment planning. We present a fully automated workflow for male pelvic CT image segmentation using deep learning. The architecture consists of a 2D organ volume localization network followed by a 3D segmentation network for volumetric segmentation of prostate, bladder, rectum, and femoral heads. We used a multi-channel 2D U-Net followed by a 3D U-Net with encoding arm modified with aggregated residual networks, known as ResNeXt. The models were trained and tested on a pelvic CT image dataset comprising 136 patients. Test results show that 3D U-Net based segmentation achieves mean (±SD) Dice coefficient values of 90 (±2.0)%, 96 (±3.0)%, 95 (±1.3)%, 95 (±1.5)%, and 84 (±3.7)% for prostate, left femoral head, right femoral head, bladder, and rectum, respectively, using the proposed fully automated segmentation method.

Automatic multiorgan segmentation in CT images of the male pelvis using region-specific hierarchical appearance cluster models

PURPOSE

Accurate segmentation of pelvic organs in CT images is of great importance in external beam radiotherapy for prostate cancer. The aim of this studying is to develop a novel method for automatic, multiorgan segmentation of the male pelvis.

METHODS

The authors' segmentation method consists of several stages. First, a pretreatment includes parameterization, principal component analysis (PCA), and an established process of region-specific hierarchical appearance cluster (RSHAC) model which was executed on the training dataset. After the preprocessing, online automatic segmentation of new CT images is achieved by combining the RSHAC model with the PCA-based point distribution model. Fifty pelvic CT from eight prostate cancer patients were used as the training dataset. From another 20 prostate cancer patients, 210 CT images were used for independent validation of the segmentation method.

RESULTS

In the training dataset, 15 PCA modes were needed to represent 95% of shape variations of pelvic organs. When tested on the validation dataset, the authors' segmentation method had an average Dice similarity coefficient and mean absolute distance of 0.751 and 0.371 cm, 0.783 and 0.303 cm, 0.573 and 0.604 cm for prostate, bladder, and rectum, respectively. The automated segmentation process took on average 5 min on a personal computer equipped with Core 2 Duo CPU of 2.8 GHz and 8 GB RAM.

CONCLUSIONS

The authors have developed an efficient and reliable method for automatic segmentation of multiple organs in the male pelvis. This method should be useful for treatment planning and adaptive replanning for prostate cancer radiotherapy. With this method, the physicist can improve the work efficiency and stability.

Anatomical medial surfaces with efficient resolution of branches singularities

Medial surfaces are powerful tools for shape description, but their use has been limited due to the sensibility of existing methods to branching artifacts. Medial branching artifacts are associated to perturbations of the object boundary rather than to geometric features. Such instability is a main obstacle for a confident application in shape recognition and description. Medial branches correspond to singularities of the medial surface and, thus, they are problematic for existing morphological and energy-based algorithms. In this paper, we use algebraic geometry concepts in an energy-based approach to compute a medial surface presenting a stable branching topology. We also present an efficient GPU-CPU implementation using standard image processing tools. We show the method computational efficiency and quality on a custom made synthetic database. Finally, we present some results on a medical imaging application for localization of abdominal pathologies.

Validation of a semiautomated liver segmentation method using CT for accurate volumetry

RATIONALE AND OBJECTIVES

To compare the repeatability and agreement of a semiautomated liver segmentation method with manual segmentation for assessment of total liver volume on CT (computed tomography).

MATERIALS AND METHODS

This retrospective, institutional review board-approved study was conducted in 41 subjects who underwent liver CT for preoperative planning. The major pathologies encountered were colorectal cancer metastases, benign liver lesions and hepatocellular carcinoma. This semiautomated segmentation method is based on variational interpolation and 3D minimal path-surface segmentation. Total and subsegmental liver volumes were segmented from contrast-enhanced CT images in venous phase. Two image analysts independently performed semiautomated segmentations and two other image analysts performed manual segmentations. Repeatability and agreement of both methods were evaluated with intraclass correlation coefficients (ICC) and Bland-Altman analysis. Interaction time was recorded for both methods.

RESULTS

Bland-Altman analysis revealed an intrareader agreement of -1 ± 27 mL (mean ± 1.96 standard deviation) with ICC of 0.999 (P < .001) for manual segmentation and 12 ± 97 mL with ICC of 0.991 (P < .001) for semiautomated segmentation. Bland-Altman analysis revealed an interreader agreement of -4 ± 22 mL with ICC of 0.999 (P < .001) for manual segmentation and 5 ± 98 mL with ICC of 0.991 (P < .001) for semiautomated segmentation. Intermethod agreement was found to be 3 ± 120 mL with ICC of 0.988 (P < .001). Mean interaction time was 34.3 ± 16.7 minutes for the manual method and 8.0 ± 1.2 minutes for the semiautomated method (P < .001).

CONCLUSIONS

A semiautomated segmentation method can substantially shorten interaction time while preserving a high repeatability and agreement with manual segmentation.

Prostate segmentation based on variant scale patch and local independent projection

Accurate segmentation of the prostate in computed tomography (CT) images is important in image-guided radiotherapy; however, difficulties remain associated with this task. In this study, an automatic framework is designed for prostate segmentation in CT images. We propose a novel image feature extraction method, namely, variant scale patch, which can provide rich image information in a low dimensional feature space. We assume that the samples from different classes lie on different nonlinear submanifolds and design a new segmentation criterion called local independent projection (LIP). In our method, a dictionary containing training samples is constructed. To utilize the latest image information, we use an online updated strategy to construct this dictionary. In the proposed LIP, locality is emphasized rather than sparsity; local anchor embedding is performed to determine the dictionary coefficients. Several morphological operations are performed to improve the achieved results. The proposed method has been evaluated based on 330 3-D images of 24 patients. Results show that the proposed method is robust and effective in segmenting prostate in CT images.

Incremental learning with selective memory (ILSM): towards fast prostate localization for image guided radiotherapy

Image-guided radiotherapy (IGRT) requires fast and accurate localization of the prostate in 3-D treatment-guided radiotherapy, which is challenging due to low tissue contrast and large anatomical variation across patients. On the other hand, the IGRT workflow involves collecting a series of computed tomography (CT) images from the same patient under treatment. These images contain valuable patient-specific information yet are often neglected by previous works. In this paper, we propose a novel learning framework, namely incremental learning with selective memory (ILSM), to effectively learn the patient-specific appearance characteristics from these patient-specific images. Specifically, starting with a population-based discriminative appearance model, ILSM aims to "personalize" the model to fit patient-specific appearance characteristics. The model is personalized with two steps: backward pruning that discards obsolete population-based knowledge and forward learning that incorporates patient-specific characteristics. By effectively combining the patient-specific characteristics with the general population statistics, the incrementally learned appearance model can localize the prostate of a specific patient much more accurately. This work has three contributions: 1) the proposed incremental learning framework can capture patient-specific characteristics more effectively, compared to traditional learning schemes, such as pure patient-specific learning, population-based learning, and mixture learning with patient-specific and population data; 2) this learning framework does not have any parametric model assumption, hence, allowing the adoption of any discriminative classifier; and 3) using ILSM, we can localize the prostate in treatment CTs accurately (DSC  ∼ 0.89 ) and fast (  ∼ 4 s), which satisfies the real-world clinical requirements of IGRT.

Overview of object oriented data analysis

Object oriented data analysis is the statistical analysis of populations of complex objects. In the special case of functional data analysis, these data objects are curves, where a variety of Euclidean approaches, such as principal components analysis, have been very successful. Challenges in modern medical image analysis motivate the statistical analysis of populations of more complex data objects that are elements of mildly non-Euclidean spaces, such as lie groups and symmetric spaces, or of strongly non-Euclidean spaces, such as spaces of tree-structured data objects. These new contexts for object oriented data analysis create several potentially large new interfaces between mathematics and statistics. The notion of object oriented data analysis also impacts data analysis, through providing a framework for discussion of the many choices needed in many modern complex data analyses, especially in interdisciplinary contexts.

Developmental changes in hippocampal shape among preadolescent children

It is known that the largest developmental changes in the hippocampus take place during the prenatal period and during the first two years of postnatal life. Few studies have been conducted to address the normal developmental trajectory of the hippocampus during childhood. In this study shape analysis was applied to study the normal developing hippocampus in a group of 103 typically developing 6- to 10-year-old preadolescent children. The individual brain was normalized to a template, and then the hippocampus was manually segmented and further divided into the head, body, and tail sub-regions. Three different methods were applied for hippocampal shape analysis: radial distance mapping, surface-based template registration using the robust point matching (RPM) algorithm, and volume-based template registration using the Demons algorithm. All three methods show that the older children have bilateral expanded head segments compared to the younger children. The results analyzed based on radial distance to the centerline were consistent with those analyzed using template-based registration methods. In analyses stratified by sex, it was found that the age-associated anatomical changes were similar in boys and girls, but the age-association was strongest in girls. Total hippocampal volume and sub-regional volumes analyzed using manual segmentation did not show a significant age-association. Our results suggest that shape analysis is sensitive to detect sub-regional differences that are not revealed in volumetric analysis. The three methods presented in this study may be applied in future studies to investigate the normal developmental trajectory of the hippocampus in children. They may be further applied to detect early deviations from the normal developmental trajectory in young children for evaluating susceptibility for psychopathological disorders involving hippocampus.

Localized differences in caudate and hippocampal shape are associated with schizophrenia but not antipsychotic type

Caudate and hippocampal volume differences in patients with schizophrenia are associated with disease and antipsychotic treatment, but local shape alterations have not been thoroughly examined. Schizophrenia patients randomly assigned to haloperidol and olanzapine treatment underwent magnetic resonance imaging (MRI) at 3, 6, and 12 months. The caudate and hippocampus were represented as medial representations (M-reps); mesh structures derived from automatic segmentations of high resolution MRIs. Two quantitative shape measures were examined: local width and local deformation. A novel nonparametric statistical method, adjusted exponentially tilted (ET) likelihood, was used to compare the shape measures across the three groups while controlling for covariates. Longitudinal shape change was not observed in the hippocampus or caudate when the treatment groups and controls were examined in a global analysis, nor when the three groups were examined individually. Both baseline and repeated measures analysis showed differences in local caudate and hippocampal size between patients and controls, while no consistent differences were shown between treatment groups. Regionally specific differences in local hippocampal and caudate shape are present in schizophrenic patients. Treatment-related related longitudinal shape change was not observed within the studied timeframe. Our results provide additional evidence for disrupted cortico-basal ganglia-thalamo-cortical circuits in schizophrenia.

CLINICAL TRIAL INFORMATION

This longitudinal study was conducted from March 1, 1997 to July 31, 2001 at 14 academic medical centers (11 in the United States, one in Canada, one in the Netherlands, and one in England). This study was performed prior to the establishment of centralized registries of federally and privately supported clinical trials.

Maximizing power to track Alzheimer's disease and MCI progression by LDA-based weighting of longitudinal ventricular surface features

We propose a new method to maximize biomarker efficiency for detecting anatomical change over time in serial MRI. Drug trials using neuroimaging become prohibitively costly if vast numbers of subjects must be assessed, so it is vital to develop efficient measures of brain change. A popular measure of efficiency is the minimal sample size (n80) needed to detect 25% change in a biomarker, with 95% confidence and 80% power. For multivariate measures of brain change, we can directly optimize n80 based on a Linear Discriminant Analysis (LDA). Here we use a supervised learning framework to optimize n80, offering two alternative solutions. With a new medial surface modeling method, we track 3D dynamic changes in the lateral ventricles in 2065 ADNI scans. We apply our LDA-based weighting to the results. Our best average n80-in two-fold nested cross-validation-is 104 MCI subjects (95% CI: [94,139]) for a 1-year drug trial, and 75AD subjects [64,102]. This compares favorably with other MRI analysis methods. The standard "statistical ROI" approach applied to the same ventricular surfaces requires 165 MCI or 94AD subjects. At 2 years, the best LDA measure needs only 67 MCI and 52AD subjects, versus 119 MCI and 80AD subjects for the stat-ROI method. Our surface-based measures are unbiased: they give no artifactual additive atrophy over three time points. Our results suggest that statistical weighting may boost efficiency of drug trials that use brain maps.

Comparison of User-Directed and Automatic Mapping of the Planned Isocenter to Treatment Space for Prostate IGRT

Image-guided radiotherapy (IGRT), adaptive radiotherapy (ART), and online reoptimization rely on accurate mapping of the radiation beam isocenter(s) from planning to treatment space. This mapping involves rigid and/or nonrigid registration of planning (pCT) and intratreatment (tCT) CT images. The purpose of this study was to retrospectively compare a fully automatic approach, including a non-rigid step, against a user-directed rigid method implemented in a clinical IGRT protocol for prostate cancer. Isocenters resulting from automatic and clinical mappings were compared to reference isocenters carefully determined in each tCT. Comparison was based on displacements from the reference isocenters and prostate dose-volume histograms (DVHs). Ten patients with a total of 243 tCTs were investigated. Fully automatic registration was found to be as accurate as the clinical protocol but more precise for all patients. The average of the unsigned x, y, and z offsets and the standard deviations (σ) of the signed offsets computed over all images were (avg. ±  σ (mm)): 1.1 ± 1.4, 1.8 ± 2.3, 2.5 ± 3.5 for the clinical protocol and 0.6 ± 0.8, 1.1 ± 1.5 and 1.1 ± 1.4 for the automatic method. No failures or outliers from automatic mapping were observed, while 8 outliers occurred for the clinical protocol.

Prostate segmentation by sparse representation based classification

PURPOSE

The segmentation of prostate in CT images is of essential importance to external beam radiotherapy, which is one of the major treatments for prostate cancer nowadays. During the radiotherapy, the prostate is radiated by high-energy x rays from different directions. In order to maximize the dose to the cancer and minimize the dose to the surrounding healthy tissues (e.g., bladder and rectum), the prostate in the new treatment image needs to be accurately localized. Therefore, the effectiveness and efficiency of external beam radiotherapy highly depend on the accurate localization of the prostate. However, due to the low contrast of the prostate with its surrounding tissues (e.g., bladder), the unpredicted prostate motion, and the large appearance variations across different treatment days, it is challenging to segment the prostate in CT images. In this paper, the authors present a novel classification based segmentation method to address these problems.

METHODS

To segment the prostate, the proposed method first uses sparse representation based classification (SRC) to enhance the prostate in CT images by pixel-wise classification, in order to overcome the limitation of poor contrast of the prostate images. Then, based on the classification results, previous segmented prostates of the same patient are used as patient-specific atlases to align onto the current treatment image and the majority voting strategy is finally adopted to segment the prostate. In order to address the limitations of the traditional SRC in pixel-wise classification, especially for the purpose of segmentation, the authors extend SRC from the following four aspects: (1) A discriminant subdictionary learning method is proposed to learn a discriminant and compact representation of training samples for each class so that the discriminant power of SRC can be increased and also SRC can be applied to the large-scale pixel-wise classification. (2) The L1 regularized sparse coding is replaced by the elastic net in order to obtain a smooth and clear prostate boundary in the classification result. (3) Residue-based linear regression is incorporated to improve the classification performance and to extend SRC from hard classification to soft classification. (4) Iterative SRC is proposed by using context information to iteratively refine the classification results.

RESULTS

The proposed method has been comprehensively evaluated on a dataset consisting of 330 CT images from 24 patients. The effectiveness of the extended SRC has been validated by comparing it with the traditional SRC based on the proposed four extensions. The experimental results show that our extended SRC can obtain not only more accurate classification results but also smoother and clearer prostate boundary than the traditional SRC. Besides, the comparison with other five state-of-the-art prostate segmentation methods indicates that our method can achieve better performance than other methods under comparison.

CONCLUSIONS

The authors have proposed a novel prostate segmentation method based on the sparse representation based classification, which can achieve considerably accurate segmentation results in CT prostate segmentation.

Development of a semi-automated method for mitral valve modeling with medial axis representation using 3D ultrasound

PURPOSE

Precise 3D modeling of the mitral valve has the potential to improve our understanding of valve morphology, particularly in the setting of mitral regurgitation (MR). Toward this goal, the authors have developed a user-initialized algorithm for reconstructing valve geometry from transesophageal 3D ultrasound (3D US) image data.

METHODS

Semi-automated image analysis was performed on transesophageal 3D US images obtained from 14 subjects with MR ranging from trace to severe. Image analysis of the mitral valve at midsystole had two stages: user-initialized segmentation and 3D deformable modeling with continuous medial representation (cm-rep). Semi-automated segmentation began with user-identification of valve location in 2D projection images generated from 3D US data. The mitral leaflets were then automatically segmented in 3D using the level set method. Second, a bileaflet deformable medial model was fitted to the binary valve segmentation by Bayesian optimization. The resulting cm-rep provided a visual reconstruction of the mitral valve, from which localized measurements of valve morphology were automatically derived. The features extracted from the fitted cm-rep included annular area, annular circumference, annular height, intercommissural width, septolateral length, total tenting volume, and percent anterior tenting volume. These measurements were compared to those obtained by expert manual tracing. Regurgitant orifice area (ROA) measurements were compared to qualitative assessments of MR severity. The accuracy of valve shape representation with cm-rep was evaluated in terms of the Dice overlap between the fitted cm-rep and its target segmentation.

RESULTS

The morphological features and anatomic ROA derived from semi-automated image analysis were consistent with manual tracing of 3D US image data and with qualitative assessments of MR severity made on clinical radiology. The fitted cm-reps accurately captured valve shape and demonstrated patient-specific differences in valve morphology among subjects with varying degrees of MR severity. Minimal variation in the Dice overlap and morphological measurements was observed when different cm-rep templates were used to initialize model fitting.

CONCLUSIONS

This study demonstrates the use of deformable medial modeling for semi-automated 3D reconstruction of mitral valve geometry using transesophageal 3D US. The proposed algorithm provides a parametric geometrical representation of the mitral leaflets, which can be used to evaluate valve morphology in clinical ultrasound images.

[Automatic image segmentation for treatment planning in radiotherapy]

One drawback of the growth in conformal radiotherapy and image-guided radiotherapy is the increased time needed to define the volumes of interest. This also results in inter- and intra-observer variability. However, developments in computing and image processing have enabled these tasks to be partially or totally automated. This article will provide a detailed description of the main principles of image segmentation in radiotherapy, its applications and the most recent results in a clinical context.

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.

Shells and Spheres: An n-Dimensional Framework for Medial-Based Image Segmentation

We have developed a method for extracting anatomical shape models from n-dimensional images using an image analysis framework we call Shells and Spheres. This framework utilizes a set of spherical operators centered at each image pixel, grown to reach, but not cross, the nearest object boundary by incorporating "shells" of pixel intensity values while analyzing intensity mean, variance, and first-order moment. Pairs of spheres on opposite sides of putative boundaries are then analyzed to determine boundary reflectance which is used to further constrain sphere size, establishing a consensus as to boundary location. The centers of a subset of spheres identified as medial (touching at least two boundaries) are connected to identify the interior of a particular anatomical structure. For the automated 3D algorithm, the only manual interaction consists of tracing a single contour on a 2D slice to optimize parameters, and identifying an initial point within the target structure.

Multi-object analysis of volume, pose, and shape using statistical discrimination

One goal of statistical shape analysis is the discrimination between two populations of objects. Whereas traditional shape analysis was mostly concerned with single objects, analysis of multi-object complexes presents new challenges related to alignment and pose. In this paper, we present a methodology for discriminant analysis of multiple objects represented by sampled medial manifolds. Non-euclidean metrics that describe geodesic distances between sets of sampled representations are used for alignment and discrimination. Our choice of discriminant method is the distance-weighted discriminant because of its generalization ability in high-dimensional, low sample size settings. Using an unbiased, soft discrimination score, we associate a statistical hypothesis test with the discrimination results. We explore the effectiveness of different choices of features as input to the discriminant analysis, using measures like volume, pose, shape, and the combination of pose and shape. Our method is applied to a longitudinal pediatric autism study with 10 subcortical brain structures in a population of 70 subjects. It is shown that the choices of type of global alignment and of intrinsic versus extrinsic shape features, the latter being sensitive to relative pose, are crucial factors for group discrimination and also for explaining the nature of shape change in terms of the application domain.

Dosimetric evaluation of automatic segmentation for adaptive IMRT for head-and-neck cancer

PURPOSE

Adaptive planning to accommodate anatomic changes during treatment requires repeat segmentation. This study uses dosimetric endpoints to assess automatically deformed contours.

METHODS AND MATERIALS

Sixteen patients with head-and-neck cancer had adaptive plans because of anatomic change during radiotherapy. Contours from the initial planning computed tomography (CT) were deformed to the mid-treatment CT using an intensity-based free-form registration algorithm then compared with the manually drawn contours for the same CT using the Dice similarity coefficient and an overlap index. The automatic contours were used to create new adaptive plans. The original and automatic adaptive plans were compared based on dosimetric outcomes of the manual contours and on plan conformality.

RESULTS

Volumes from the manual and automatic segmentation were similar; only the gross tumor volume (GTV) was significantly different. Automatic plans achieved lower mean coverage for the GTV: V95: 98.6 +/- 1.9% vs. 89.9 +/- 10.1% (p = 0.004) and clinical target volume: V95: 98.4 +/- 0.8% vs. 89.8 +/- 6.2% (p < 0.001) and a higher mean maximum dose to 1 cm(3) of the spinal cord 39.9 +/- 3.7 Gy vs. 42.8 +/- 5.4 Gy (p = 0.034), but no difference for the remaining structures.

CONCLUSIONS

Automatic segmentation is not robust enough to substitute for physician-drawn volumes, particularly for the GTV. However, it generates normal structure contours of sufficient accuracy when assessed by dosimetric end points.

Image estimation from marker locations for dose calculation in prostate radiation therapy

Tracking implanted markers in the prostate during each radiation treatment delivery provides an accurate approximation of prostate location, which enables the use of higher daily doses with tighter margins of the treatment beams and thus improves the efficiency of the radiotherapy. However, the lack of 3D image data with such a technique prevents calculation of delivered dose as required for adaptive planning. We propose to use a reference statistical shape model generated from the planning image and a deformed version of the reference model fitted to the implanted marker locations during treatment to estimate a regionally dense deformation from the planning space to the treatment space. Our method provides a means of estimating the treatment image by mapping planning image data to treatment space via the deformation field and therefore enables the calculation of dose distributions with marker tracking techniques during each treatment delivery.

Combining a deformable model and a probabilistic framework for an automatic 3D segmentation of prostate on MRI

PURPOSE

Accurate localization and contouring of prostate are crucial issues in prostate cancer diagnosis and/or therapies. Although several semi-automatic and automatic segmentation methods have been proposed, manual expert correction remains necessary. We introduce a new method for automatic 3D segmentation of the prostate gland from magnetic resonance imaging (MRI) scans.

METHODS

A statistical shape model was used as an a priori knowledge, and gray levels distribution was modeled by fitting histogram modes with a Gaussian mixture. Markov fields were used to introduce contextual information regarding voxels' neighborhoods. Final labeling optimization is based on Bayesian a posteriori classification, estimated with the iterative conditional mode algorithm.

RESULTS

We compared the accuracy of this method, free from any manual correction, with contours outlined by an expert radiologist. In 12 cases, including prostates with cancer and benign prostatic hypertrophy, the mean Hausdorff distance and overlap ratio were 9.94 mm and 0.83, respectively.

CONCLUSION

This new automatic prostate MRI segmentation method produces satisfactory results, even at prostate's base and apex. The method is computationally feasible and efficient.

Continuous medial representation of brain structures using the biharmonic PDE

A new approach for constructing deformable continuous medial models for anatomical structures is presented. Medial models describe geometrical objects by first specifying the skeleton of the object and then deriving the boundary surface corresponding to the skeleton. However, an arbitrary specification of a skeleton will not be "valid" unless a certain set of sufficient conditions is satisfied. The most challenging of these is the non-linear equality constraint that must hold along the boundaries of the manifolds forming the skeleton. The main contribution of this paper is to leverage the biharmonic partial differential equation as a mapping from a codimension-0 subset of Euclidean space to the space of skeletons that satisfy the equality constraint. The PDE supports robust numerical solution on freeform triangular meshes, providing additional flexibility for shape modeling. The approach is evaluated by generating continuous medial models for a large dataset of hippocampus shapes. Generalizations to modeling more complex shapes and to representing branching skeletons are demonstrated.

Training models of anatomic shape variability

Learning probability distributions of the shape of anatomic structures requires fitting shape representations to human expert segmentations from training sets of medical images. The quality of statistical segmentation and registration methods is directly related to the quality of this initial shape fitting, yet the subject is largely overlooked or described in an ad hoc way. This article presents a set of general principles to guide such training. Our novel method is to jointly estimate both the best geometric model for any given image and the shape distribution for the entire population of training images by iteratively relaxing purely geometric constraints in favor of the converging shape probabilities as the fitted objects converge to their target segmentations. The geometric constraints are carefully crafted both to obtain legal, nonself-interpenetrating shapes and to impose the model-to-model correspondences required for useful statistical analysis. The paper closes with example applications of the method to synthetic and real patient CT image sets, including same patient male pelvis and head and neck images, and cross patient kidney and brain images. Finally, we outline how this shape training serves as the basis for our approach to IGRT/ART.

Development of a novel post-processing treatment planning platform for 4D radiotherapy

The aim of this study is to develop an Automatic Post-processing Tool for four-dimensional (4D) treatment planning (APT4D) that enables the user to perform some necessary procedures related to 4D treatment planning, such as automated image registration, automatic propagation of regions of interest, and dose distribution transformation. Demons-based deformable registrations were performed to map the moving phase images (such as the end-inhalation phase or 0% phase) to the reference phase (typically the end-exhalation fixed phase or 50% phase). Contours were automatically propagated into the moving phase using the image registration results. The dose distribution of each moving phase was transformed to the fixed phase and subsequently was summed as an average with equal weighting factor. To validate the application of APT4D utility, the 4D computed tomography (CT) images of a lung cancer patient and an abdominal cancer patient were acquired and resorted into ten respiratory phases. 4D plans based on the 4D CT images were developed. The correlation coefficient ranged from 0.992 to 0.999 for the re-sampled deformed moving phase image against the fixed phase image for the lung patient plan and from 0.977 to 0.999 for the abdominal patient plan. For all the organs, the match indices between the manual contours and automatic contour propagation results were around 0.92 to 0.95. The 4D composite dose-volume histogram showed dosimetric reductions for liver and kidneys in the high dose region. The APT4D adds automation, efficiency, and functionality, while integrating the whole process of 4D treatment planning.

Concurrent multimodality image segmentation by active contours for radiotherapy treatment planning

Multimodality imaging information is regularly used now in radiotherapy treatment planning for cancer patients. The authors are investigating methods to take advantage of all the imaging information available for joint target registration and segmentation, including multimodality images or multiple image sets from the same modality. In particular, the authors have developed variational methods based on multivalued level set deformable models for simultaneous 2D or 3D segmentation of multimodality images consisting of combinations of coregistered PET, CT, or MR data sets. The combined information is integrated to define the overall biophysical structure volume. The authors demonstrate the methods on three patient data sets, including a nonsmall cell lung cancer case with PET/CT, a cervix cancer case with PET/CT, and a prostate patient case with CT and MRI. CT, PET, and MR phantom data were also used for quantitative validation of the proposed multimodality segmentation approach. The corresponding Dice similarity coefficient (DSC) was 0.90 +/- 0.02 (p < 0.0001) with an estimated target volume error of 1.28 +/- 1.23% volume. Preliminary results indicate that concurrent multimodality segmentation methods can provide a feasible and accurate framework for combining imaging data from different modalities and are potentially useful tools for the delineation of biophysical structure volumes in radiotherapy treatment planning.

[Dosimetric evaluation of an automatic segmentation tool of pelvic structures from MRI images for prostate cancer radiotherapy]

PURPOSE

An automatic segmentation tool of pelvic structures from MRI images for prostate cancer radiotherapy was developed and dosimetric evaluation of differences of delineation (automatic versus human) is presented here.

MATERIALS AND METHODS

CTV, rectum and bladder were defined automatically and by a physician in 20 patients. Treatment plans based on "automatic" volumes were transferred on "manual" volumes and reciprocally. Dosimetric characteristics of PTV (V(95), minimal, maximal and mean doses), rectum (V(50), V(70), maximal and mean doses) and bladder (V(70), maximal and mean doses) were compared.

RESULTS

Automatic delineation of CTV did not significantly influence dosimetric characteristics of "manual" PTV. Rectal V(50) and V(70) were not significantly different; mean rectal dose is slightly superior (43.2 versus 44.4Gy, p=0.02, Student test). Bladder V(70) was significantly superior too (19.3 versus 21.6, p=0.004). Organ-at-risk (OAR) automatic delineation had little influence on their dosimetric characteristics; rectal V(70) was slightly underestimated (20 versus 18.5Gy, p=0.001).

CONCLUSION

CTV and OAR automatic delineation had little influence on dosimetric characteristics. Software developments are ongoing to enable routine use and interobserver evaluation is needed.

Automatic segmentation of bladder and prostate using coupled 3D deformable models

In this paper, we propose a fully automatic method for the coupled 3D localization and segmentation of lower abdomen structures. We apply it to the joint segmentation of the prostate and bladder in a database of CT scans of the lower abdomen of male patients. A flexible approach on the bladder allows the process to easily adapt to high shape variation and to intensity inhomogeneities that would be hard to characterize (due, for example, to the level of contrast agent that is present). On the other hand, a statistical shape prior is enforced on the prostate. We also propose an adaptive non-overlapping constraint that arbitrates the evolution of both structures based on the availability of strong image data at their common boundary. The method has been tested on a database of 16 volumetric images, and the validation process includes an assessment of inter-expert variability in prostate delineation, with promising results.

Automated finite-element analysis for deformable registration of prostate images

Two major factors preventing the routine clinical use of finite-element analysis for image registration are: 1) the substantial labor required to construct a finite-element model for an individual patient's anatomy and 2) the difficulty of determining an appropriate set of finite-element boundary conditions. This paper addresses these issues by presenting algorithms that automatically generate a high quality hexahedral finite-element mesh and automatically calculate boundary conditions for an imaged patient. Medial shape models called m-reps are used to facilitate these tasks and reduce the effort required to apply finite-element analysis to image registration. Encouraging results are presented for the registration of CT image pairs which exhibit deformation caused by pressure from an endorectal imaging probe and deformation due to swelling.

Automatic Segmentation of Pelvic Structures From Magnetic Resonance Images for Prostate Cancer Radiotherapy

PURPOSE

Target-volume and organ-at-risk delineation is a time-consuming task in radiotherapy planning. The development of automated segmentation tools remains problematic, because of pelvic organ shape variability. We evaluate a three-dimensional (3D), deformable-model approach and a seeded region-growing algorithm for automatic delineation of the prostate and organs-at-risk on magnetic resonance images.

METHODS AND MATERIALS

Manual and automatic delineation were compared in 24 patients using a sagittal T2-weighted (T2-w) turbo spin echo (TSE) sequence and an axial T1-weighted (T1-w) 3D fast-field echo (FFE) or TSE sequence. For automatic prostate delineation, an organ model-based method was used. Prostates without seminal vesicles were delineated as the clinical target volume (CTV). For automatic bladder and rectum delineation, a seeded region-growing method was used. Manual contouring was considered the reference method. The following parameters were measured: volume ratio (Vr) (automatic/manual), volume overlap (Vo) (ratio of the volume of intersection to the volume of union; optimal value = 1), and correctly delineated volume (Vc) (percent ratio of the volume of intersection to the manually defined volume; optimal value = 100).

RESULTS

For the CTV, the Vr, Vo, and Vc were 1.13 (+/-0.1 SD), 0.78 (+/-0.05 SD), and 94.75 (+/-3.3 SD), respectively. For the rectum, the Vr, Vo, and Vc were 0.97 (+/-0.1 SD), 0.78 (+/-0.06 SD), and 86.52 (+/-5 SD), respectively. For the bladder, the Vr, Vo, and Vc were 0.95 (+/-0.03 SD), 0.88 (+/-0.03 SD), and 91.29 (+/-3.1 SD), respectively.

CONCLUSIONS

Our results show that the organ-model method is robust, and results in reproducible prostate segmentation with minor interactive corrections. For automatic bladder and rectum delineation, magnetic resonance imaging soft-tissue contrast enables the use of region-growing methods.

Assessment of a Model-Based Deformable Image Registration Approach for Radiation Therapy Planning

PURPOSE

The aim of this study is to develop a surface-based deformable image registration strategy and to assess the accuracy of the system for the integration of multimodality imaging, image-guided radiation therapy, and assessment of geometrical change during and after therapy.

METHODS AND MATERIALS

A surface-model-based deformable image registration system has been developed that enables quantitative description of geometrical change in multimodal images with high computational efficiency. Based on the deformation of organ surfaces, a volumetric deformation field is derived using different volumetric elasticity models as alternatives to finite-element modeling.

RESULTS

The accuracy of the system was assessed both visually and quantitatively by tracking naturally occurring landmarks (bronchial bifurcations in the lung, vessel bifurcations in the liver, implanted gold markers in the prostate). The maximum displacements for lung, liver and prostate were 5.3 cm, 3.2 cm, and 0.6 cm respectively. The largest registration error (direction, mean +/- SD) for lung, liver and prostate were (inferior-superior, -0.21 +/- 0.38 cm), (anterior-posterior, -0.09 +/- 0.34 cm), and (left-right, 0.04 +/- 0.38 cm) respectively, which was within the image resolution regardless of the deformation model. The computation time (2.7 GHz Intel Xeon) was on the order of seconds (e.g., 10 s for 2 prostate datasets), and deformed axial images could be viewed at interactive speed (less than 1 s for 512 x 512 voxels).

CONCLUSIONS

Surface-based deformable image registration enables the quantification of geometrical change in normal tissue and tumor with acceptable accuracy and speed.

Multi-figure Anatomical Objects for Shape Statistics

Multi-figure m-reps allow us to represent and analyze a complex anatomical object by its parts, by relations among its parts, and by the object itself as a whole entity. This representation also enables us to gather either global or hierarchical statistics from a population of such objects. We propose a framework to train the statistics of multi-figure anatomical objects from real patient data. This training requires fitting multi-figure m-reps to binary characteristic images of training objects. To evaluate the fitting approach, we propose a Monte Carlo method sampling the trained statistics. It shows that our methods generate geometrically proper models that are close to the set of Monte Carlo generated target models and thus can be expected to yield similar statistics to that used for the Monte Carlo generation.