Advanced High-Quality Maximum Intensity Projection for Volume Visualization Lukas Mroz, Helwig L¨ offelmann, and Eduard Gr¨ oller Vienna University of Technology, Institute of Computer Graphics Abstract Maximum Intensity Projection (MIP) is a volume rendering tech- nique which is used to extract high-intensity structures from volu- metric data. At each pixel the highest data value encountered along the corresponding viewing ray is determined. MIP is commonly used to extract vascular structures from medical MRI data sets (an- giography). Due to lack of depth information in MIP images, ani- mation of the viewpoint is frequently used for viewing. Although interactive MIP algorithms exist, the quality of the results is mod- erate. The generation of high-quality MIP animation loops is com- putationally expensive with rendering times of several seconds per frame. In this paper we present a fast algorithm for high-quality MIP. Cells of the data set which will never contribute to a MIP due to their neighborhood are removed during a preprocessing step. The remaining cells are stored in a way which improves cache co- herency independent of the viewing direction and minimizes the number of required maximum evaluations. keywords: Volume Visualization, Maximum Intensity Projec- tion, angiography 1 Introduction The ability to depict blood vessels is of great importance for many medical imaging applications. CT and MRI scanners can be used to obtain volumetric data sets which allow the extraction of vascular structures. Especially data originating from MRI, which are most frequently used for this purpose, exhibit some properties which make the application of standard volume visualization techniques like ray casting [4] or iso-surface extraction [6] difficult. MRI data sets contain a significant amount of noise. Inhomogeneities in the sampled data make it difficult to extract surfaces of objects by spec- ifying a single iso-value. In addition vascular structures and espe- cially thin vessels cover a wide range of data-values making the extraction by conventional techniques difficult. Maximum intensity projection exploits the fact, that within an- giography data sets the data values of vascular structures are higher than the values of the surrounding tissue. By depicting the maxi- mum data value seen through each pixel, the structure of the vessels contained in the data is captured. A straightforward method for cal- culating MIP is to perform ray casting and search for the maximum sample value along the ray instead of the usual opacity and color compositing process done in volume rendering. In contrast to di- rect volume rendering, no early ray termination is possible and the whole volume has to be processed. As a MIP contains no shading information, depth and occlusion information is lost. Structures with higher data value lying behind a lower valued object even appear to be in front of it. The most com- mon way to ease the interpretation of such images is to animate the viewpoint while viewing. This can be achieved using one of the interactive MIP techniques [1, 3, 7]. For performance reasons these Institute of Computer Graphics, Vienna University of Technology, Karlsplatz 13/186/2, A-1040 Vienna, Austria. emal: mroz, loeffelmann, groeller @cg.tuwien.ac.at techniques perform no resampling of the original data values dur- ing maximum evaluation, thus introducing aliasing and producing images of just moderate quality. For exact depiction of even small features there is need for algorithms which produce high-quality MIP-animations in reasonable time. In contrast to interactive MIP techniques which perform no resampling, more accurate evaluation of ray maxima is required for the generation of high-quality MIP. At the cost of significantly longer computation times, this allows to create much more detailed images and animations with a higher resolution. Depending on the quality requirements of the resulting image and desired performance, different strategies for finding the maxi- mum value along a ray can be used. Analytical solution: Usually the data values within a cell are assumed to result from trilinear interpolation of the cell ver- tices. For each data cell which is intersected by the ray the maximum value encountered by the ray is calculated analyti- cally. As the data value along a ray through a cell is described by a cubic polynomial, this means finding a maximum con- tained within a section of the curve. This is the most accurate but also computationally most expensive method [8]. Sampling and interpolation: As usually done for ray cast- ing, data values are sampled along the ray using trilinear in- terpolation. The cost of this approach depends on the step size along the ray and on how many interpolations that do not affect the result can be avoided [8, 9]. Nearest neighbor interpolation: Values of the data points closest to the ray are taken for maximum estimation. In com- bination with discrete ray traversal this is the fastest method. As no interpolation is performed, the voxel structure is visible in the resulting image as aliasing [1]. Recent algorithms for MIP employ a set of approaches for speeding up the rendering: Ray traversal and interpolation optimization: Sakas et al. [8] evaluate cell maxima only if the maximum value of the examined cell is larger than the ray-maximum calculated so far. For additional speedup they use integer arithmetic for ray traversal and a cache-coherent volume storage scheme. Zuiderveld et al. [9] apply a similar technique to avoid tri- linear interpolations. In addition, cells containing only back- ground noise are not interpolated. For further speedup a low- resolution image containing lower-bound maximum estima- tions for pixel clusters is generated before the main rendering step. Cells with values below this bound can be skipped when the final image is generated. Finally a distance-volume is used to skip empty spaces. A careful definition of the condition for “empty space” is required to avoid missing small, low valued structures. Use graphics hardware: Heidrich at al. [3] use conven- tional polygon rendering hardware to simulate MIP. Several iso-surfaces for different threshold values are extracted from