Fucang Jia

Cloud-Based Evaluation of Anatomical Structure Segmentation and Landmark Detection Algorithms: VISCERAL Anatomy Benchmarks

Variations in the shape and appearance of anatomical structures in medical images are often relevant radiological signs of disease. Automatic tools can help automate parts of this manual process. A cloud-based evaluation framework is presented in this paper including results of benchmarking current state-of-the-art medical imaging algorithms for anatomical structure segmentation and landmark detection: the VISCERAL Anatomy benchmarks. The algorithms are implemented in virtual machines in the cloud where participants can only access the training data and can be run privately by the benchmark administrators to objectively compare their performance in an unseen common test set. Overall, 120 computed tomography and magnetic resonance patient volumes were manually annotated to create a standard Gold Corpus containing a total of 1295 structures and 1760 landmarks. Ten participants contributed with automatic algorithms for the organ segmentation task, and three for the landmark localization task. Different algorithms obtained the best scores in the four available imaging modalities and for subsets of anatomical structures. The annotation framework, resulting data set, evaluation setup, results and performance analysis from the three VISCERAL Anatomy benchmarks are presented in this article. Both the VISCERAL data set and Silver Corpus generated with the fusion of the participant algorithms on a larger set of non-manually-annotated medical images are available to the research community.

Segmentation of Brain Tissues from Magnetic Resonance Images Using Adaptively Regularized Kernel-Based Fuzzy C-Means Clustering.

An adaptively regularized kernel-based fuzzy C-means clustering framework is proposed for segmentation of brain magnetic resonance images. The framework can be in the form of three algorithms for the local average grayscale being replaced by the grayscale of the average filter, median filter, and devised weighted images, respectively. The algorithms employ the heterogeneity of grayscales in the neighborhood and exploit this measure for local contextual information and replace the standard Euclidean distance with Gaussian radial basis kernel functions. The main advantages are adaptiveness to local context, enhanced robustness to preserve image details, independence of clustering parameters, and decreased computational costs. The algorithms have been validated against both synthetic and clinical magnetic resonance images with different types and levels of noises and compared with 6 recent soft clustering algorithms. Experimental results show that the proposed algorithms are superior in preserving image details and segmentation accuracy while maintaining a low computational complexity.

Brain tumor segmentation from multimodal magnetic resonance images via sparse representation

OBJECTIVE Accurately segmenting and quantifying brain gliomas from magnetic resonance (MR) images remains a challenging task because of the large spatial and structural variability among brain tumors. To develop a fully automatic and accurate brain tumor segmentation algorithm, we present a probabilistic model of multimodal MR brain tumor segmentation. This model combines sparse representation and the Markov random field (MRF) to solve the spatial and structural variability problem. METHODS We formulate the tumor segmentation problem as a multi-classification task by labeling each voxel as the maximum posterior probability. We estimate the maximum a posteriori (MAP) probability by introducing the sparse representation into a likelihood probability and a MRF into the prior probability. Considering the MAP as an NP-hard problem, we convert the maximum posterior probability estimation into a minimum energy optimization problem and employ graph cuts to find the solution to the MAP estimation. RESULTS Our method is evaluated using the Brain Tumor Segmentation Challenge 2013 database (BRATS 2013) and obtained Dice coefficient metric values of 0.85, 0.75, and 0.69 on the high-grade Challenge data set, 0.73, 0.56, and 0.54 on the high-grade Challenge LeaderBoard data set, and 0.84, 0.54, and 0.57 on the low-grade Challenge data set for the complete, core, and enhancing regions. CONCLUSIONS The experimental results show that the proposed algorithm is valid and ranks 2nd compared with the state-of-the-art tumor segmentation algorithms in the MICCAI BRATS 2013 challenge.

Scale-adaptive surface modeling of vascular structures

BackgroundThe effective geometric modeling of vascular structures is crucial for diagnosis, therapy planning and medical education. These applications require good balance with respect to surface smoothness, surface accuracy, triangle quality and surface size.MethodsOur method first extracts the vascular boundary voxels from the segmentation result, and utilizes these voxels to build a three-dimensional (3D) point cloud whose normal vectors are estimated via covariance analysis. Then a 3D implicit indicator function is computed from the oriented 3D point cloud by solving a Poisson equation. Finally the vessel surface is generated by a proposed adaptive polygonization algorithm for explicit 3D visualization.ResultsExperiments carried out on several typical vascular structures demonstrate that the presented method yields both a smooth morphologically correct and a topologically preserved two-manifold surface, which is scale-adaptive to the local curvature of the surface. Furthermore, the presented method produces fewer and better-shaped triangles with satisfactory surface quality and accuracy.ConclusionsCompared to other state-of-the-art approaches, our method reaches good balance in terms of smoothness, accuracy, triangle quality and surface size. The vessel surfaces produced by our method are suitable for applications such as computational fluid dynamics simulations and real-time virtual interventional surgery.

A marker-based watershed method for X-ray image segmentation

Digital X-ray images are the most frequent modality for both screening and diagnosis in hospitals. To facilitate subsequent analysis such as quantification and computer aided diagnosis (CAD), it is desirable to exclude image background. A marker-based watershed segmentation method was proposed to segment background of X-ray images. The method consisted of six modules: image preprocessing, gradient computation, marker extraction, watershed segmentation from markers, region merging and background extraction. One hundred clinical direct radiograph X-ray images were used to validate the method. Manual thresholding and multiscale gradient based watershed method were implemented for comparison. The proposed method yielded a dice coefficient of 0.964±0.069, which was better than that of the manual thresholding (0.937±0.119) and that of multiscale gradient based watershed method (0.942±0.098). Special means were adopted to decrease the computational cost, including getting rid of few pixels with highest grayscale via percentile, calculation of gradient magnitude through simple operations, decreasing the number of markers by appropriate thresholding, and merging regions based on simple grayscale statistics. As a result, the processing time was at most 6s even for a 3072×3072 image on a Pentium 4 PC with 2.4GHz CPU (4 cores) and 2G RAM, which was more than one time faster than that of the multiscale gradient based watershed method. The proposed method could be a potential tool for diagnosis and quantification of X-ray images.

Automatic subarachnoid space segmentation and hemorrhage detection in clinical head CT scans

PurposeThe subarachnoid space (SAS) lies between the arachnoid membrane and the pia mater of the human brain, normally filled with cerebrospinal fluid (CSF). Subarachnoid hemorrhage (SAH) is a serious complication of neurological disease that can have high mortality and high risk of disability. Computed tomography (CT) head scans are often used for diagnosing SAH which may be difficult when the hemorrhage is small or subtle. A computer-aided diagnosis system from CT images is thus developed to augment image interpretation.MethodsSupervised learning using the probability of distance features of several landmarks was employed to recognize SAS. For each CT image, the SAS was approximated in four steps: (1) Landmarks including brain boundary, midsagittal plane (MSP), anterior and posterior intersection points of brain boundary with the MSP, and superior point of the brain were extracted. (2) Distances to all the landmarks were calculated for every pixel in the CT image, and combined to construct a high-dimensional feature vector. (3) Using head CT images with manually delineated SAS as training dataset, the prior probabilities of distances for pixels within SAS and non-SAS were computed. (4) Any pixel of a head CT scan in the testing dataset was classified as an SAS or non-SAS pixel in a Bayesian decision framework based on its distance features.ResultsThe proposed method was validated on clinical head CT images by comparison with manual segmentation. The results showed that the automated method is consistent with the gold standard. Compared with elastic registration based on grayscale information, the proposed method was less affected by grayscale variation between normal controls and patients. Compared with manual delineation, the average spatial overlap, relative overlap, and similarity index were, respectively, 89, 63, and 76% for the automatic SAS approximation of the 69 head CT scans tested. The proposed method was tested for SAH detection and yielded a sensitivity of 100% and a specificity of 92%.ConclusionAutomated SAH detection with high sensitivity was shown feasible in a prototype computer-aided diagnosis system. The proposed method may be extended for computer-aided diagnosis of several CSF-related diseases relevant to SAS abnormalities.

The development of the intrinsic functional connectivity of default network subsystems from age 3 to 5

In recent years, research on human functional brain imaging using resting-state fMRI techniques has been increasingly prevalent. The term “default mode” was proposed to describe a baseline or default state of the brain during rest. Recent studies suggested that the default mode network (DMN) is comprised of two functionally distinct subsystems: a dorsal-medial prefrontal cortex (DMPFC) subsystem involved in self-oriented cognition (i.e., theory of mind) and a medial temporal lobe (MTL) subsystem engaged in memory and scene construction; both subsystems interact with the anterior medial prefrontal cortex (aMPFC) and posterior cingulate (PCC) as the core regions of DMN. The present study explored the development of DMN core regions and these two subsystems in both hemispheres from 3- to 5-year-old children. The analysis of the intrinsic activity showed strong developmental changes in both subsystems, and significant changes were specifically found in MTL subsystem, but not in DMPFC subsystem, implying distinct developmental trajectories for DMN subsystems. We found stronger interactions between the DMPFC and MTL subsystems in 5-year-olds, particularly in the left subsystems that support the development of environmental adaptation and relatively complex mental activities. These results also indicate that there is stronger right hemispheric lateralization at age 3, which then changes as bilateral development gradually increases through to age 5, suggesting in turn the hemispheric dominance in DMN subsystems changing with age. The present results provide primary evidence for the development of DMN subsystems in early life, which might be closely related to the development of social cognition in childhood.

Automatic Liver Segmentation Using Statistical Prior Models and Free-form Deformation

In this paper, an automatic and robust coarse-to-fine liver image segmentation method is proposed. Multiple prior knowledge models are built to implement liver localization and segmentation: voxel-based AdaBoost classifier is trained to localize liver position robustly, shape and appearance models are constructed to fit liver these models to original CT volume. Free-form deformation is incorporated to improve the models’ ability of refining liver boundary. The method was submitted to VISCERAL big data challenge, and had been tested on IBSI 2014 challenge datasets and the result demonstrates that the proposed method is accurate and efficient.

Curvature-dependent surface visualization of vascular structures

Efficient visualization of vascular structures is essential for therapy planning and medical education. Existing techniques achieve high-quality visualization of vascular surfaces at the cost of low rendering speed and large size of resulting surface. In this paper, we present an approach for visualizing vascular structures by exploiting the local curvature information of a given surface. To handle complex topology of loop and multiple parents and/or multiple children, bidirectional adaptive sampling and modified normal calculations at joints are proposed. The proposed method has been applied to cerebral vascular trees, liver vessel trees, and aortic vessel trees. The experimental results show that it can obtain a high-quality surface visualization with fewer polygons in the approximation.

Fast automatic 3D liver segmentation based on a three‐level AdaBoost‐guided active shape model

PURPOSE A robust, automatic, and rapid method for liver delineation is urgently needed for the diagnosis and treatment of liver disorders. Until now, the high variability in liver shape, local image artifacts, and the presence of tumors have complicated the development of automatic 3D liver segmentation. In this study, an automatic three-level AdaBoost-guided active shape model (ASM) is proposed for the segmentation of the liver based on enhanced computed tomography images in a robust and fast manner, with an emphasis on the detection of tumors. METHODS The AdaBoost voxel classifier and AdaBoost profile classifier were used to automatically guide three-level active shape modeling. In the first level of model initialization, fast automatic liver segmentation by an AdaBoost voxel classifier method is proposed. A shape model is then initialized by registration with the resulting rough segmentation. In the second level of active shape model fitting, a prior model based on the two-class AdaBoost profile classifier is proposed to identify the optimal surface. In the third level, a deformable simplex mesh with profile probability and curvature constraint as the external force is used to refine the shape fitting result. In total, three registration methods-3D similarity registration, probability atlas B-spline, and their proposed deformable closest point registration-are used to establish shape correspondence. RESULTS The proposed method was evaluated using three public challenge datasets: 3Dircadb1, SLIVER07, and Visceral Anatomy3. The results showed that our approach performs with promising efficiency, with an average of 35 s, and accuracy, with an average Dice similarity coefficient (DSC) of 0.94 ± 0.02, 0.96 ± 0.01, and 0.94 ± 0.02 for the 3Dircadb1, SLIVER07, and Anatomy3 training datasets, respectively. The DSC of the SLIVER07 testing and Anatomy3 unseen testing datasets were 0.964 and 0.933, respectively. CONCLUSIONS The proposed automatic approach achieves robust, accurate, and fast liver segmentation for 3D CTce datasets. The AdaBoost voxel classifier can detect liver area quickly without errors and provides sufficient liver shape information for model initialization. The AdaBoost profile classifier achieves sufficient accuracy and greatly decreases segmentation time. These results show that the proposed segmentation method achieves a level of accuracy comparable to that of state-of-the-art automatic methods based on ASM.