Active contour segmentation and its robust implementation using level set methods are well established theoretical approaches that have been studied thoroughly in the image analysis literature. Despite the existence of these powerful segmentation methods, the needs of clinical research continue to be fulfilled, to a large extent, using slice-by-slice manual tracing. To bridge the gap between methodological advances and clinical routine, we developed an open source application called ITK-SNAP, which is intended to make level set segmentation easily accessible to a wide range of users, including those with little or no mathematical expertise. This paper describes the methods and software engineering philosophy behind this new tool and provides the results of validation experiments performed in the context of an ongoing child autism neuroimaging study. The validation establishes SNAP intra/interrater reliability and overlap error statistics for the caudate nucleus and finds that SNAP is a highly reliable and efficient alternative to manual tracing.Analogous results for lateral ventricle segmentation are provided.

No abstract

Abstract-This paper discusses the development of a new method for the automatic segmentation of anatomical structures from volumetric medical images. Driving application is the segmentation of 3-D tumor structures from magnetic resonance images (MRI), which is known to be a very challenging segmentation problem due to the variability of tumor geometry and intensity patterns. Level set evolution combining global smoothness with the flexibility of topology changes offers significant advantages over conventional statistical classification followed by mathematical morphology. Level set evolution with constant propagation needs to be initialized either completely inside or outside and can leak through weak or missing boundary parts. Replacing the constant propagation term by a signed local statistical force overcomes these limitations and results in a region competition method that converges to a stable solution.Applied to MR images presenting tumors, probabilities for background and tumor regions are calculated from a preand post-contrast difference image and mixture-modelling fit of the histogram. The whole image is used for initialization of the level set evolution to segment the blobby-shaped tumor boundaries. Preliminary results on five cases presenting different tumors with significant shape and intensity variability demonstrate that the new method might become a powerful and efficient tool for the clinic. Validity is demonstrated by comparison with manual expert segmentation.

Abstract. Given models for healthy brains, tumor segmentation can be seen as a process of detecting abnormalities or outliers that are present with certain image intensity and geometric properties. In this paper, we propose a method that segments brain tumor and edema in two stages. We first detect intensity outliers using robust estimation of the location and dispersion of the normal brain tissue intensity clusters. We then apply geometric and spatial constraints to the detected abnormalities or outliers. Previously published tumor segmentation methods generally rely on the intensity enhancement in the T1-weighted image that appear with the gadolinium contrast agent, on strictly uniform intensity patterns and most often on user initialization of the segmentation. To our knowledge, none of the methods integrated the detection of edema in addition to tumor as a combined approach, although knowledge of the extent of edema is critical for planning and treatment. Our method relies on the information provided by the (non-enhancing) T1 and T2 image channels, the use of a registered probabilistic brain atlas as a spatial prior, and the use of a shape prior for the tumor/edema region. The result is an efficient, automatic segmentation method that defines both, tumor and edema.

This paper presents results for real-time visualization of out-of-core collections of 3D objects. This is a significant extension of previous methods and shows the generality of hierarchical paging procedures applied both to global terrain and any objects that reside on it. Applied to buildings, the procedure shows the effectiveness of using a screen-based paging and display criterion within a hierarchical framework. The results demonstrate that the method is scalable since it is able to handle multiple collections of buildings (e.g., cities) placed around the earth with full interactivity and without extensive memory load. Further the method shows efficient handling of culling and is applicable to larger, extended collections of buildings. Finally, the method shows that levels of detail can be incorporated to provide improved detail management.

Abstract. We face the question of how to produce a scale space of image intensities relative to a scale space of objects or other characteristic image regions filling up the image space, when both images and objects are understood to come from a population. We argue for a schema combining a multi-scale image representation with a multi-scale representation of objects or regions. The objects or regions at one scale level are produced using soft-edged apertures, which are subdivided into sub-regions. The intensities in the regions are represented using histograms. Relevant probabilities of region shape and inter-relations between region geometry and of histograms are described, and the means is given of interrelating the intensity probabilities and geometric probabilities by producing the probabilities of intensities conditioned on geometry.

Abstract. We present a novel statistical image-match model for use in Bayesian segmentation, a multiscale extension of image profile models akin to those in Active Shape Models. A spherical-harmonic based 3D shape representation provides a mapping of the object boundary to the sphere S 2 , and a scale-space for profiles on the sphere defines a scalespace on the object. A key feature is that profiles are not blurred across the object boundary, but only along the boundary. This profile scalespace is sampled in a coarse-to-fine fashion to produce features for the statistical image-match model. A framework for model-building and segmentation has been built, and testing and validation are in progress with a dataset of 70 segmented images of the caudate nucleus. Why Are Anatomical Objects So Hard to Segment?Model-based segmentation has come a long way since Kass and Witkin's original snakes [1], but segmentation of anatomical structures from real-world 3D medical images still presents some difficult challenges for automatic methods. Bayesian model-based segmentation balances a geometry prior, guided by a model of the expected object shape, against an image match likelihood, guided by a model of the expected image appearance around the object. Much has been done on the shape representation and prior; here we will focus on the image match model. In many objects, a simple gradient-magnitude based image-match model is insufficient. The profile of the image across the object boundary can vary significantly from one portion of the boundary to another. Some portions of the boundary might not even have a visible contrast, in which case the shape prior is needed to define the contour. In real-world medical images, the contrast-to-noise ratio is often low, and models need to be robust to image noise.In our study, one of the applications we focus on is the caudate nucleus in the human brain. From a partnership with our Psychiatry department, we have access to over 70 high-resolution MRIs (T1-weighted, 1x1x1mm) with highquality manual expert segmentations of both left and right caudates. The manual raters, having spent much effort on developing a reliable protocol for manual segmentation, indicate some of the challenges in caudate segmentation, which motivate a multiscale statistical image-match model for automatic methods.

Active contour segmentation and its robust implementation using level sets have been studied thoroughly in the medical image analysis literature. Despite the availability of these powerful methods, clinical research still largely relies on manual slice-by-slice outlining for anatomical structure segmentation. To bridge the gap between methodological advances and clinical routine, we developed ITK-SNAP: an open source application intended to make level set segmentation easily accessible to a wide range of users with various levels of mathematical expertise. We briefly describe this new tool and report the results of a validation study in which ITK-SNAP was compared to manual segmentation of the caudate in the context of an ongoing child neuroimaging autism study.