Naeem Shareef

Segmentation of medical images using LEGION

Advances in visualization technology and specialized graphic workstations allow clinicians to virtually interact with anatomical structures contained within sampled medical-image datasets. A hindrance to the effective use of this technology is the difficult problem of image segmentation. In this paper, the authors utilize a recently proposed oscillator network called the locally excitatory globally inhibitory oscillator network (LEGION) whose ability to achieve fast synchrony with local excitation and desynchrony with global inhibition makes it an effective computational framework for grouping similar features and segregating dissimilar ones in an image. The authors extract an algorithm from LEGION dynamics and propose an adaptive scheme for grouping. They show results of the algorithm to two-dimensional (2-D) and three-dimensional (3-D) (volume) computerized topography (CT) and magnetic resonance imaging (MRI) medical-image datasets. In addition, the authors compare their algorithm with other algorithms for medical-image segmentation, as well as with manual segmentation. LEGIONs computational and architectural properties make it a promising approach for real-time medical-image segmentation.

Hardware assisted volume rendering of unstructured grids by incremental slicing

Some of the more important research results in computational science rely on the use of simulation methods that operate on unstructured grids. However, these grids, composed of a set of polyhedra, introduce exceptional problems with respect to data visualization. Volume rendering techniques, originally developed to handle rectangular grids, show significant promise for general use with unstructured grids as well. The main disadvantage of this approach, compared to isosurfaces, particles or other visualization tools is its non-interactive performance. We describe an efficient method for rendering unstructured grids that is based on incremental slicing and hardware polygon rendering. For a given view direction, the grid vertices are transformed to image space using available graphics hardware. We then incrementally compute the 2D polygon-meshes that result from letting a set of planes, parallel to the screen plane, intersect (slice) the transformed grid. Finally, we use the graphics hardware to render (interpolate-fill) the polygon-meshes and composite them in visibility order. We show that, in addition to being faster than existing methods, our approach also provides adaptive control and progressive image generation. The adaptive method provides user control to ensure that the contribution of every cell is included in the final image or to limit the number of cells that are missed.

High-quality splatting on rectilinear grids with efficient culling of occluded voxels

Splatting is a popular volume rendering algorithm that pairs good image quality with an efficient volume projection scheme. The current axis-aligned sheet-buffer approach, however, bears certain inaccuracies. The effect of these is less noticeable in still images, but clearly revealed in animated viewing, where disturbing popping of object brightness occurs at certain view angle transitions. In previous work, we presented a new variant of sheet-buffered splatting in which the compositing sheets are oriented parallel to the image plane. This scheme not only eliminates the condition for popping, but also produces images of higher quality. In this paper, we summarize this new paradigm and extend it in a number of ways. We devise a new solution to render rectilinear grids of equivalent cost to the traditional approach that treats the anisotropic volume as being warped into a cubic grid. This enables us to use the usual radially symmetric kernels, which can be projected without inaccuracies. Next, current splatting approaches necessitate the projection of all voxels in the iso-interval(s), although only a subset of these voxels may eventually be visible in the final image. To eliminate these wasteful computations we propose a novel front-to-back approach that employs an occlusion map to determine if a splat contributes to the image before it is projected, thus skipping occluded splats. Additional measures are presented for further speedups. In addition, we present an efficient list-based volume traversal scheme that facilitates the quick modification of transfer functions and iso-values.

Splatting errors and antialiasing

The paper describes three new results for volume rendering algorithms utilizing splatting. First, an antialiasing extension to the basic splatting algorithm is introduced that mitigates the spatial aliasing for high resolution volumes. Aliasing can be severe for high resolution volumes or volumes where a high depth of field leads to converging samples along the perspective axis. Next, an analysis of the common approximation errors in the splatting process for perspective viewing is presented. In this context, we give different implementations, distinguished by efficiency and accuracy, for adding the splat contributions to the image plane. We then present new results in controlling the splatting errors and also show their behavior in the framework of our new antialiasing technique. Finally, current work in progress on extensions to splatting for temporal antialiasing is demonstrated. We present a simple but highly effective scheme for adding motion blur to fast moving volumes.

An anti-aliasing technique for splatting

Splatting is a popular direct volume rendering algorithm. However, the algorithm does not correctly render cases where the volume sampling rate is higher than the image sampling rate (e.g. more than one voxel maps into a pixel). This situation arises with orthographic projections of high-resolution volumes, as well as with perspective projections of volumes of any resolution. The result is potentially severe spatial and temporal aliasing artifacts. Some volume ray-casting algorithms avoid these artifacts by employing reconstruction kernels which vary in width as the rays diverge. Unlike ray-casting algorithms, existing splatting algorithms do not have an equivalent mechanism for avoiding these artifacts. The authors propose such a mechanism, which delivers high-quality splatted images and has the potential for a very efficient hardware implementation.

FastSplats: optimized splatting on rectilinear grids

Splatting is widely applied in many areas, including volume, point-based and image-based rendering. Improvements to splatting, such as eliminating popping and color bleeding, occasion-based acceleration, post-rendering classification and shading, have all been recently accomplished. These improvements share a common need for efficient frame-buffer accesses. We present an optimized software splatting package, using a newly designed primitive, called FastSplat, to scan-convert footprints. Our approach does not use texture mapping hardware, but supports the whole pipeline in memory. In such an integrated pipeline, we are then able to study the optimization strategies and address image quality issues. While this research is meant for a study of the inherent trade-off of splatting, our renderer, purely in software, achieves 3- to 5-fold speedups over a top-end texture hardware implementation (for opaque data sets). We further propose a method of efficient occlusion culling using a summed area table of opacity. 3D solid texturing and bump mapping capabilities are demonstrated to show the flexibility of such an integrated rendering pipeline. A detailed numerical error analysis, in addition to the performance and storage issues, is also presented. Our approach requires low storage and uses simple operations. Thus, it is easily implementable in hardware.

Rapid previewing via volume-based solid modeling

Quick previewing of 3D models is necessary for efficient product design and rapid prototyping. An inherent weakness of most solid modeling systems is that as model complexity increases, quick 3D viewing suffers, especially on low cost workstations. We explore an alternative approach to surface representation in which object space, called volume, is subdivided into a 3D grid of cubic cells, each containing information on the object(s) which occupy it. A data structure is introduced that consists of a 2D amay of pointers each holding a linked list of adjacent non empty cells. To benefit from data coherency along one dimension, we have developed new modeling and rendering algorithms that are beam-oriented, incremental, and integer-based. To illustrate the usefulness of our approach, we use it in Constructive Solid Geometry (CSG) modeling. We describe our prototype system and show, by comparing it to existing systems, that our data structure and its associated algorithms, while being of finite resolution, provide for suitable and more efficient model visualization.

View-dependent approach to MIP for very large data

A simple and yet useful approach to visualize a variety of structures from sampled data is the Maximum Intensity Projection (MIP). Higher valued structures of interest pass in the projection over occluding structures. This can make MIP images difficult to interpret due to the loss of depth information. Animating about the data is one key way to try to decipher such ambiguities. The challenge is that MIP is inherently expensive and thus high frame rates are difficult to achieve. Variations to the original MIP algorithm and classification can help to further alleviate ambiguities and also provide improved image quality and very different visualizations. But they make the technique even more expensive. In addition, they require much parameter searching and tweaking. As todays data sizes are increasingly getting larger, current methods only allow very limited interaction. We explore a view-dependent approach using concepts from image-based rendering. A novel multi-layered image representation storing scalar information is computed at a view sample and then warped to the users view. We present algorithms using OpenGL to quickly compute MIP and its variations using commodity off-the-shelf graphics hardware to achieve near interactive rates.

An Image-Based Modelling Approach To GPU-based Unstructured Grid Volume Rendering

Unstructured grid volume rendering on the GPU is an ongoing research goal. The main difficulty is how to map the irregular data efficiently to the graphics hardware. Grid cells cannot be directly processed by the pipeline since polyhedral cells are not basic graphics primitives. Thus, approaches that render directly from the grid must try to overcome inherent computational bottlenecks such as cell ordering. We present a novel image-based approach that allows for flexible sample placement and an efficient mapping to graphics hardware that can leverage current trends in hardware performance. The unstructured grid is replaced by a semi-regular data structure we call the Pixel Ray Image (PRI), which consists of a pixelized projection plane and samples defined on sampling rays that intersect the grid. Each ray is interpreted as a linear spline of scalar values. The knot count is reduced by taking advantage of coherence in the sampling direction using spline simplification. The final representation is stored in a packed fashion onto 2D textures and uploaded once to texture memory prior to rendering. We detail how to perform efficient rendering utilizing shader programming and a slice-based rendering scheme that is similar to those used by hardware-accelerated rendering approaches for regular grids.