Volume Data Visualization
by Alexey Tukalo
Volumetric data is very common nowadays. The importance of this dataset type will grow rapidly due to the development of the 3D data acquisition field, and the possibilities to perform an advanced visualization on a modern ofﬁce workstation with an interactive framerate.
The dataset can be captured by various technologies, e.g. MRI, CT, PET, USCT or echolocation. It also can be produced by physical simulations, for example, ﬂuid dynamics or particle systems. The set of technologies mentioned before demonstrates that volumetric information plays an important role in medicine. It is used for an advanced cancer detection, visualization of aneurysms, and treatment planning. This kind of data is also very useful for non-destructive material testing via computer tomography or ultrasound. In addition, huge three-dimensional dataset is produced by geoseismic research.
Overview of Volume Data
Volume data consists of voxels. Voxel is the basic volume element. It can be represented as a point in a 3D space with a certain position and a color. This gives an opportunity to keep up to six scalar parameters.
Usually the points belong to the fixed grid, so the volume data can be stored as a table. In this case the runtime representation can be kept as a multidimensional array, and volume data can be represented as a *.csv file in local storage. However, more commonly the dataset is broken into several slices, and every slice is stored as a bitmap image. The approach allows to significantly reduce the model size due to the sophisticated compressing algorithm which could be applied to the images.
Volume data visualization
There are four main ways to visualize the volumetric dataset. Below we are going to discuss the advantages and disadvantages of the technologies.
This is the most straight-forward solution, which implies the separate visualization of every slice of the volume dataset with an opportunity to scroll them interactively.
The simplicity of the implementation and low computational complexity are the key advantages of the technique. However, its main problem is that the viewer should use his/her imagination to reconstruct an entire object structure. As the result, the slice-based approach is not the most suitable for visual analyses of very complex and unknown structures. But it suits well for the detection of features inside the well-known objects, such as parts of human body. That is why the methodology is widely used in medicine. For example, it is the most popular way of representation for MRI and CT. It is worth to mention, that general CT and MRI studies have much lower resolution in one of the dimensions, which causes some difficulties for utilizing the datasets with more advanced technologies.
Emulation of Other Technologies
This approach can be very useful when visual analyses should be performed by experts, who are accustomed to the certain technology. For example, it can be used during the development of new technologies in medical and geoseismic areas. The emulation allows experts to have a smooth transition to a modern technic from their old solutions. The approach is not very popular due to several reasons. First, it requires the usage of the very detailed volume dataset, but the major part of the information would be lost or spoiled by mimicking another technology. For this reason, the popularity of the visualization will be steadily decreasing during the integration of the new technology to the experts’ workflow. Secondly, the development of this visualization type requires too much time to archive the results which would be close to the initial technology’s images, and its usage can be dropped in the end of the transition period. An additional problem of the approach is that it requires certain knowledge and experience for the correct results interpretation.
Visualization of 3D object as a 2D image is called 3D rendering. The most common way of the 3D rendering is based on photo-realistic visualization of surfaces which are represented by polygonal meshes. The technology is utilized so widely that modern graphic card’s architecture is designed to accelerate the operation.
Indirect Volume Rendering
Since there are many tools for visualizing polygonal mesh models, the idea behind Indirect Volume Rendering is obvious. The approach consists of two steps. The first one is an extraction of isosurface out of a dataset in accordance with certain threshold. There are several algorithms exist to perform the task. The most popular one is Marching Cubes. Sometimes isosurface extraction can be improved by developing the special algorithm which is based on a specific feature of the particular dataset. Then the polygonal surface model can be visualized by any 3D engine or other tools for visualization of polygonal mesh models, e.g. LightningChart MeshModels.
The main advantages of the approach are inherited from the technology behind. It contains all the typical features of 3D object visualization such as rotation, usage of different amounts of light sources, interaction with other 3D objects and so on. As the result, it makes complex 3D structures analysis much simpler. It is especially useful for visual detection of meaningful details inside the unknown datasets. Due to the performance optimization of common 3D rendering engines, the visualization can be handled by any modern office workstation. In addition, the technique allows developers to use much more sophisticated noise reduction algorithms.
The main disadvantages of the approach are caused by the first step of visualization process. Firstly, conversion of the volume dataset to the polygonal mesh surface leads to the loss of the data from inside the surface area. In addition, the isosurface extraction algorithm can require complex calculations, so preprocessing can take noticeable amount of time, that is why usually it is impossible to interactively change the threshold of the surface extraction.
Direct Volume Rendering
Direct Volume Rendering does not require any preprocessing. The data is visualized from an original dataset. It gives the algorithms an opportunity to modify the transfer function and threshold dynamically. Also, some of the approaches allow to visualize the internal structure of the dataset in semi-transparent way.
Worth to mention that it is practically the most powerful way to visualize volume data. The visualization has all the advantages of polygonal mesh models, and can be easily combined with them on the same scene. In addition, it is possible to cut a part of the model for an investigation of structures hidden by the object surface.
The main disadvantage of the approach was a high-hardware requirement, but due to modern graphic card development achievements, it is already possible to run the visualization even with the cheapest options. Another problem is the high cost of the own volume rendering engine development.
There are several different technological implementations of Direct Volume Rendering. The most common of them use the tools created for a GPU acceleration of polygonal mesh models rendering in their own way. Texture-based Volume and Volume Ray Casting are presently the most successful approaches of direct volume rendering.
Texture-based Volume Rendering technic uses a set of planes to construct the object. The dataset is projected to the planes as textures. The final picture is combined by alpha blending of the planes. Volume Ray Casting approach uses cube as a placeholder for the volume model. The model itself is projected to the sides of the cube by the Ray Casting algorithm which uses rays to accumulate the data and combine it with the specific equation called Ray Function.
Ray Function is truly fascinating feature of Volume Ray Casting. It allows to define how rays perform sampling of the dataset and calculation of the pixel color. Different Ray Functions can extract different features out of the dataset. Let’s discuss three examples of Ray Function:
Accumulation Function tries to collect and combine as much data as possible to give a viewer an opportunity to explore an internal structure of the object. The visualization produced by this technique looks like a semi-transparent gel.
Accumulation Ray Function application visualizing the medical dataset.
Maximum Intensity Function visualizes only the brightest value sampled by the ray. Visually it provides similar result to the X-ray images. It allows to get an additional information about the internal structure of the object.
Maximum Intensity Function application for the ultrasound waves interference simulation.
Isosurface Rendering draws the model surface in a way that it looks like polygonal model rendering. The final result is very similar to those produced by the Indirect Volume Rendering.
Isosurface Ray Function application for visualization of water flow simulation.
The development of hardware prepares the base for the growth of interest in different volume data acquisition technologies. Improvements in consumer’s computer performance will have a positive effect on popularity of advanced volume visualization techniques, such as Direct and Indirect Volume Rendering, while more basic ones will become less popular.
LightningChart has an outstanding tool for the implementation of Slice-based and Indirect Volume rendering visualizations of volume data. Moreover, our own Direct Volume Rendering Engine with a lot of advanced features for volume data visualization will be available soon.
The database of University of Iowa was utilised to provide the images of Volume Data visualization in this article.