Evaluation and comparison of current fetal ultrasound image segmentation methods for biometric measurements: a grand challenge.

This paper presents the evaluation results of the methods submitted to Challenge US: Biometric Measurements from Fetal Ultrasound Images, a segmentation challenge held at the IEEE International Symposium on Biomedical Imaging 2012. The challenge was set to compare and evaluate current fetal ultrasound image segmentation methods. It consisted of automatically segmenting fetal anatomical structures to measure standard obstetric biometric parameters, from 2D fetal ultrasound images taken on fetuses at different gestational ages (21 weeks, 28 weeks, and 33 weeks) and with varying image quality to reflect data encountered in real clinical environments. Four independent sub-challenges were proposed, according to the objects of interest measured in clinical practice: abdomen, head, femur, and whole fetus. Five teams participated in the head sub-challenge and two teams in the femur sub-challenge, including one team who tackled both. Nobody attempted the abdomen and whole fetus sub-challenges. The challenge goals were two-fold and the participants were asked to submit the segmentation results as well as the measurements derived from the segmented objects. Extensive quantitative (region-based, distance-based, and Bland-Altman measurements) and qualitative evaluation was performed to compare the results from a representative selection of current methods submitted to the challenge. Several experts (three for the head sub-challenge and two for the femur sub-challenge), with different degrees of expertise, manually delineated the objects of interest to define the ground truth used within the evaluation framework. For the head sub-challenge, several groups produced results that could be potentially used in clinical settings, with comparable performance to manual delineations. The femur sub-challenge had inferior performance to the head sub-challenge due to the fact that it is a harder segmentation problem and that the techniques presented relied more on the femur's appearance.

Multimodal evaluation for medical image segmentation

This paper is a joint effort between five institutionsthat introduces several novel similarity measures andcombines them to carry out a multimodal segmentationevaluation. The new similarity measures proposed arebased on the location and the intensity values of themisclassified voxels as well as on the connectivity andthe boundaries of the segmented data. We showexperimentally that the combination of these measuresimprove the quality of the evaluation. The study that weshow here has been carried out using four differentsegmentation methods from four different labs applied toa MRI simulated dataset of the brain. We claim that ournew measures improve the robustness of the evaluation andprovides better understanding about the differencebetween segmentation methods.

A multidimensional segmentation evaluation for medical image data.

Evaluation of segmentation methods is a crucial aspect in image processing, especially in the medical imaging field, where small differences between segmented regions in the anatomy can be of paramount importance. Usually, segmentation evaluation is based on a measure that depends on the number of segmented voxels inside and outside of some reference regions that are called gold standards. Although some other measures have been also used, in this work we propose a set of new similarity measures, based on different features, such as the location and intensity values of the misclassified voxels, and the connectivity and the boundaries of the segmented data. Using the multidimensional information provided by these measures, we propose a new evaluation method whose results are visualized applying a Principal Component Analysis of the data, obtaining a simplified graphical method to compare different segmentation results. We have carried out an intensive study using several classic segmentation methods applied to a set of MRI simulated data of the brain with several noise and RF inhomogeneity levels, and also to real data, showing that the new measures proposed here and the results that we have obtained from the multidimensional evaluation, improve the robustness of the evaluation and provides better understanding about the difference between segmentation methods.

An Efficient Segmentation Method for Ultrasound Images based on a Semi-supervised Approach and Patch-based Features

Segmenting ultrasound images is a challenging problemwhere standard unsupervised segmentation methods such asthe well-known Chan-Vese method fail. We propose in thispaper an efficient segmentation method for this class ofimages. Our proposed algorithm is based on asemi-supervised approach (user labels) and the use ofimage patches as data features. We also consider thePearson distance between patches, which has been shown tobe robust w.r.t speckle noise present in ultrasoundimages. Our results on phantom and clinical data show avery high similarity agreement with the ground truthprovided by a medical expert.

Active deformation fields: dense deformation field estimation for atlas-based segmentation using the active contour framework.

This paper presents a new and original variational framework for atlas-based segmentation. The proposed framework integrates both the active contour framework, and the dense deformation fields of optical flow framework. This framework is quite general and encompasses many of the state-of-the-art atlas-based segmentation methods. It also allows to perform the registration of atlas and target images based on only selected structures of interest. The versatility and potentiality of the proposed framework are demonstrated by presenting three diverse applications: In the first application, we show how the proposed framework can be used to simulate the growth of inconsistent structures like a tumor in an atlas. In the second application, we estimate the position of nonvisible brain structures based on the surrounding structures and validate the results by comparing with other methods. In the final application, we present the segmentation of lymph nodes in the Head and Neck CT images, and demonstrate how multiple registration forces can be used in this framework in an hierarchical manner.

Harmonic active contours.

We propose a segmentation method based on the geometric representation of images as 2-D manifolds embedded in a higher dimensional space. The segmentation is formulated as a minimization problem, where the contours are described by a level set function and the objective functional corresponds to the surface of the image manifold. In this geometric framework, both data-fidelity and regularity terms of the segmentation are represented by a single functional that intrinsically aligns the gradients of the level set function with the gradients of the image and results in a segmentation criterion that exploits the directional information of image gradients to overcome image inhomogeneities and fragmented contours. The proposed formulation combines this robust alignment of gradients with attractive properties of previous methods developed in the same geometric framework: 1) the natural coupling of image channels proposed for anisotropic diffusion and 2) the ability of subjective surfaces to detect weak edges and close fragmented boundaries. The potential of such a geometric approach lies in the general definition of Riemannian manifolds, which naturally generalizes existing segmentation methods (the geodesic active contours, the active contours without edges, and the robust edge integrator) to higher dimensional spaces, non-flat images, and feature spaces. Our experiments show that the proposed technique improves the segmentation of multi-channel images, images subject to inhomogeneities, and images characterized by geometric structures like ridges or valleys.

Central and Cortical Gray Mater Segmentation of Magnetic Resonance Images of the Fetal Brain

Motivation. The study of human brain development in itsearly stage is today possible thanks to in vivo fetalmagnetic resonance imaging (MRI) techniques. Aquantitative analysis of fetal cortical surfacerepresents a new approach which can be used as a markerof the cerebral maturation (as gyration) and also forstudying central nervous system pathologies [1]. However,this quantitative approach is a major challenge forseveral reasons. First, movement of the fetus inside theamniotic cavity requires very fast MRI sequences tominimize motion artifacts, resulting in a poor spatialresolution and/or lower SNR. Second, due to the ongoingmyelination and cortical maturation, the appearance ofthe developing brain differs very much from thehomogenous tissue types found in adults. Third, due tolow resolution, fetal MR images considerably suffer ofpartial volume (PV) effect, sometimes in large areas.Today extensive efforts are made to deal with thereconstruction of high resolution 3D fetal volumes[2,3,4] to cope with intra-volume motion and low SNR.However, few studies exist related to the automatedsegmentation of MR fetal imaging. [5] and [6] work on thesegmentation of specific areas of the fetal brain such asposterior fossa, brainstem or germinal matrix. Firstattempt for automated brain tissue segmentation has beenpresented in [7] and in our previous work [8]. Bothmethods apply the Expectation-Maximization Markov RandomField (EM-MRF) framework but contrary to [7] we do notneed from any anatomical atlas prior. Data set &Methods. Prenatal MR imaging was performed with a 1-Tsystem (GE Medical Systems, Milwaukee) using single shotfast spin echo (ssFSE) sequences (TR 7000 ms, TE 180 ms,FOV 40 x 40 cm, slice thickness 5.4mm, in plane spatialresolution 1.09mm). Each fetus has 6 axial volumes(around 15 slices per volume), each of them acquired inabout 1 min. Each volume is shifted by 1 mm with respectto the previous one. Gestational age (GA) ranges from 29to 32 weeks. Mother is under sedation. Each volume ismanually segmented to extract fetal brain fromsurrounding maternal tissues. Then, in-homogeneityintensity correction is performed using [9] and linearintensity normalization is performed to have intensityvalues that range from 0 to 255. Note that due tointra-tissue variability of developing brain someintensity variability still remains. For each fetus, ahigh spatial resolution image of isotropic voxel size of1.09 mm is created applying [2] and using B-splines forthe scattered data interpolation [10] (see Fig. 1). Then,basal ganglia (BS) segmentation is performed on thissuper reconstructed volume. Active contour framework witha Level Set (LS) implementation is used. Our LS follows aslightly different formulation from well-known Chan-Vese[11] formulation. In our case, the LS evolves forcing themean of the inside of the curve to be the mean intensityof basal ganglia. Moreover, we add local spatial priorthrough a probabilistic map created by fitting anellipsoid onto the basal ganglia region. Some userinteraction is needed to set the mean intensity of BG(green dots in Fig. 2) and the initial fitting points forthe probabilistic prior map (blue points in Fig. 2). Oncebasal ganglia are removed from the image, brain tissuesegmentation is performed as described in [8]. Results.The case study presented here has 29 weeks of GA. Thehigh resolution reconstructed volume is presented in Fig.1. The steps of BG segmentation are shown in Fig. 2.Overlap in comparison with manual segmentation isquantified by the Dice similarity index (DSI) equal to0.829 (values above 0.7 are considered a very goodagreement). Such BG segmentation has been applied on 3other subjects ranging for 29 to 32 GA and the DSI hasbeen of 0.856, 0.794 and 0.785. Our segmentation of theinner (red and blue contours) and outer cortical surface(green contour) is presented in Fig. 3. Finally, torefine the results we include our WM segmentation in theFreesurfer software [12] and some manual corrections toobtain Fig.4. Discussion. Precise cortical surfaceextraction of fetal brain is needed for quantitativestudies of early human brain development. Our workcombines the well known statistical classificationframework with the active contour segmentation forcentral gray mater extraction. A main advantage of thepresented procedure for fetal brain surface extraction isthat we do not include any spatial prior coming fromanatomical atlases. The results presented here arepreliminary but promising. Our efforts are now in testingsuch approach on a wider range of gestational ages thatwe will include in the final version of this work andstudying as well its generalization to different scannersand different type of MRI sequences. References. [1]Guibaud, Prenatal Diagnosis 29(4) (2009). [2] Rousseau,Acad. Rad. 13(9), 2006, [3] Jiang, IEEE TMI 2007. [4]Warfield IADB, MICCAI 2009. [5] Claude, IEEE Trans. Bio.Eng. 51(4) (2004). [6] Habas, MICCAI (Pt. 1) 2008. [7]Bertelsen, ISMRM 2009 [8] Bach Cuadra, IADB, MICCAI 2009.[9] Styner, IEEE TMI 19(39 (2000). [10] Lee, IEEE Trans.Visual. And Comp. Graph. 3(3), 1997, [11] Chan, IEEETrans. Img. Proc, 10(2), 2001 [12] Freesurfer,http://surfer.nmr.mgh.harvard.edu.

Semi-supervised segmentation of ultrasound images based on patch representation and continuous min cut.

Ultrasound segmentation is a challenging problem due to the inherent speckle and some artifacts like shadows, attenuation and signal dropout. Existing methods need to include strong priors like shape priors or analytical intensity models to succeed in the segmentation. However, such priors tend to limit these methods to a specific target or imaging settings, and they are not always applicable to pathological cases. This work introduces a semi-supervised segmentation framework for ultrasound imaging that alleviates the limitation of fully automatic segmentation, that is, it is applicable to any kind of target and imaging settings. Our methodology uses a graph of image patches to represent the ultrasound image and user-assisted initialization with labels, which acts as soft priors. The segmentation problem is formulated as a continuous minimum cut problem and solved with an efficient optimization algorithm. We validate our segmentation framework on clinical ultrasound imaging (prostate, fetus, and tumors of the liver and eye). We obtain high similarity agreement with the ground truth provided by medical expert delineations in all applications (94% DICE values in average) and the proposed algorithm performs favorably with the literature.

Representing diffusion MRI in 5D for segmentation of white matter tracts with a level set method.

We present a method for segmenting white matter tracts from high angular resolution diffusion MR. images by representing the data in a 5 dimensional space of position and orientation. Whereas crossing fiber tracts cannot be separated in 3D position space, they clearly disentangle in 5D position-orientation space. The segmentation is done using a 5D level set method applied to hyper-surfaces evolving in 5D position-orientation space. In this paper we present a methodology for constructing the position-orientation space. We then show how to implement the standard level set method in such a non-Euclidean high dimensional space. The level set theory is basically defined for N-dimensions but there are several practical implementation details to consider, such as mean curvature. Finally, we will show results from a synthetic model and a few preliminary results on real data of a human brain acquired by high angular resolution diffusion MRI.

Segmentation and grid generation for numerical simulations of vad connections with patient-specific data

Objectives: We are interested in the numerical simulation of the anastomotic region comprised between outflow canula of LVAD and the aorta. Segmenta¬tion, geometry reconstruction and grid generation from patient-specific data remain an issue because of the variable quality of DICOM images, in particular CT-scan (e.g. metallic noise of the device, non-aortic contrast phase). We pro¬pose a general framework to overcome this problem and create suitable grids for numerical simulations.Methods: Preliminary treatment of images is performed by reducing the level window and enhancing the contrast of the greyscale image using contrast-limited adaptive histogram equalization. A gradient anisotropic diffusion filter is applied to reduce the noise. Then, watershed segmentation algorithms and mathematical morphology filters allow reconstructing the patient geometry. This is done using the InsightToolKit library (www.itk.org). Finally the Vascular Model¬ing ToolKit (www.vmtk.org) and gmsh (www.geuz.org/gmsh) are used to create the meshes for the fluid (blood) and structure (arterial wall, outflow canula) and to a priori identify the boundary layers. The method is tested on five different patients with left ventricular assistance and who underwent a CT-scan exam.Results: This method produced good results in four patients. The anastomosis area is recovered and the generated grids are suitable for numerical simulations. In one patient the method failed to produce a good segmentation because of the small dimension of the aortic arch with respect to the image resolution.Conclusions: The described framework allows the use of data that could not be otherwise segmented by standard automatic segmentation tools. In particular the computational grids that have been generated are suitable for simulations that take into account fluid-structure interactions. Finally the presented method features a good reproducibility and fast application.

Weighted Shape-based Averaging with Neighborhood Prior Model for Multiple Atlas Fusion-based Medical Image Segmentation

In medical imaging, merging automated segmentations obtained from multiple atlases has become a standard practice for improving the accuracy. In this letter, we propose two new fusion methods: "Global Weighted Shape-Based Averaging" (GWSBA) and "Local Weighted Shape-Based Averaging" (LWSBA). These methods extend the well known Shape-Based Averaging (SBA) by additionally incorporating the similarity information between the reference (i.e., atlas) images and the target image to be segmented. We also propose a new spatially-varying similarity-weighted neighborhood prior model, and an edge-preserving smoothness term that can be used with many of the existing fusion methods. We first present our new Markov Random Field (MRF) based fusion framework that models the above mentioned information. The proposed methods are evaluated in the context of segmentation of lymph nodes in the head and neck 3D CT images, and they resulted in more accurate segmentations compared to the existing SBA.

MP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-mapping at high field.

The large spatial inhomogeneity in transmit B(1) field (B(1)(+)) observable in human MR images at high static magnetic fields (B(0)) severely impairs image quality. To overcome this effect in brain T(1)-weighted images, the MPRAGE sequence was modified to generate two different images at different inversion times, MP2RAGE. By combining the two images in a novel fashion, it was possible to create T(1)-weighted images where the result image was free of proton density contrast, T(2) contrast, reception bias field, and, to first order, transmit field inhomogeneity. MP2RAGE sequence parameters were optimized using Bloch equations to maximize contrast-to-noise ratio per unit of time between brain tissues and minimize the effect of B(1)(+) variations through space. Images of high anatomical quality and excellent brain tissue differentiation suitable for applications such as segmentation and voxel-based morphometry were obtained at 3 and 7 T. From such T(1)-weighted images, acquired within 12 min, high-resolution 3D T(1) maps were routinely calculated at 7 T with sub-millimeter voxel resolution (0.65-0.85 mm isotropic). T(1) maps were validated in phantom experiments. In humans, the T(1) values obtained at 7 T were 1.15+/-0.06 s for white matter (WM) and 1.92+/-0.16 s for grey matter (GM), in good agreement with literature values obtained at lower spatial resolution. At 3 T, where whole-brain acquisitions with 1 mm isotropic voxels were acquired in 8 min, the T(1) values obtained (0.81+/-0.03 s for WM and 1.35+/-0.05 for GM) were once again found to be in very good agreement with values in the literature.

Comparison of Energy Minimization Methods for 3-D Brain Tissue Classification

This paper presents 3-D brain tissue classificationschemes using three recent promising energy minimizationmethods for Markov random fields: graph cuts, loopybelief propagation and tree-reweighted message passing.The classification is performed using the well knownfinite Gaussian mixture Markov Random Field model.Results from the above methods are compared with widelyused iterative conditional modes algorithm. Theevaluation is performed on a dataset containing simulatedT1-weighted MR brain volumes with varying noise andintensity non-uniformities. The comparisons are performedin terms of energies as well as based on ground truthsegmentations, using various quantitative metrics.