The 2007 vizNET Showcase prize winners were awarded as follows:
First Prize Second Prize
Even in most advanced modelling studies carried out today, the structure of the heart is oversimplified, and major determinants of the cardiac activation sequence like the specialized conduction system, fiber orientation, or the complex endocardial structures of cardiac cavities are largely missing . Among the available cardiac imaging modalities, MRI and histological sectioning have been used for generation of cardiac models. While the resolution and tissue discrimination capabilities of histological images are unparalleled, it is a destructive technique and suffers from slice-to-slice misalignment and non-linear deformations introduced in the preparation process, which are not present in the MRI images. The scientific goal is to combine histology and MRI images of the whole heart, in order to obtain a high-resolution, detailed, anatomically correct cardiac model
Rabbit hearts were isolated and mounted to a Langendorff perfusion system. Hearts were fixed with minimal delay during cardioplegic arrest, and then embedded in wax. Whole hearts were serially sectioned (10 μm thickness). Sections were Trichrome stained to identify collagen (bluish green), myocytes (pink), cytoplasm (orange, highlighting non-myocytes), and nuclei. Stained sections were mounted and imaged to obtain whole cardiac cross section mosaic images, yielding a final image resolution of 5.4 μm. Before serial sectioning, the hearts were MRI scanned. Imaging was carried out on an 11.7 T (500 MHz) MR system. For high resolution gap-free 3D MRI, fixed hearts were scanned with an in-plane resolution of 26.4 μm × 26.4 μm, and an out-of plane resolution of 24.4 μm.
The computational work towards achieving the scientific aim presents a method for establishing 3D correspondence between MRI and histological datasets to obtain a highly detailed, geometrically correct anatomical description of the heart. Histological slices provide unique resolution and ability to discriminate between different tissues and anatomical structures; however, they suffer from slice misalignment and deformation. MRI datasets preserve the correct geometry but provide less micro-structural detail. An iterative process is used to correct the various 2D and 3D, rigid and non-rigid deformations. Work is in progress to exploit grid middleware and distributed server side visualisation for computation and visualisation. The resulting histology volume can be used to map cell types and fibre orientation on MRI, which can be used to generate detailed cardiac models. These computationally expensive problems require Grid Computing for parallel processing and visualisation. Figure 3 below represents a simple overview of how the authors have used a commodity hardware and open source software stack based visualization cluster located at Rutherford Appleton Laboratory to facilitate scientists situated in Oxford to visualise the high resolution image data in near real time.
The visualization environment is a turn-key cluster graphics system based on high-end computing hardware and enhanced Chromium software. Its render server is an open source engine for remote and collaborative visualization. The STFC cluster has 17 dual processor nodes with NVIDIA Quadro FX3450 card and 4GB memory on each node.
The aim was to see if an off the shelf visualization software toolkit can be used with the hardware to support the visualization of all this data. The rationale was that such toolkits use will speed up the take up of remote visualization, especially if there is a severe limitation on the amount of time and effort available for developing tailored solutions. There is also an interest to explore how the visualization environment catered to the use of existing visualization tools which users may have familiarity with.
In the first instance, ParaView was used on a single node but that supported the visualization of only a small portion of the data. The use of 8 nodes in the ParaView client server model was found sufficient for rendering surfaces from half the data in near real time. A useful built-in functionality was that the system automatically used a low resolution version of the displayed image when the user interacted with the image, for instance, rotated it. When the user let go of the mouse, the system took a few seconds, under 5, to render the high resolution image. Volume visualization of the whole dataset using ParaView was found impossible with the given resources. This is mostly due to how volume visualization is handled by ParaView. In addition, the support automatic smoothing of surfaces in ParaView was not good enough.
Raptor, a Chromium based renderer, provided high resolution volume rendering more naturally. It loads the data into the graphics card’s texture memory for rendering. It was possible to render a quarter of the data using Raptor with 11 processors in real time. The limitation is due to the graphics card which only supported 512x512 textures. The default partitioning used by Raptor is to partition the data in the z direction, along slices. It is possible to overcome this limitation by re-implementing the partition algorithm or by using cards with more texture memory. In conclusion, it can be said that ParaView is a visualization toolkit that supported varied functionality, including parallel rendering. However, it was not possible to get good performance. Raptor gave better performance but is of limited functionality.
The experimental biologists and the computing and visualization team came together for the first time to undertake this experiment in a very short period of time. That in itself engendered interesting sociological interactions, especially as the data is precious and of very high value and there was little time to establish trusted relationships. It also became apparent that at least some members of the biologists team were users of VTK suite of software and felt comfortable with its use. However, the collaboration engendered a useful shared experience and set the ground for ongoing collaboration to turn the application truly parallelised end to end and implement the data transformation and image processing algorithms more generically.
Flying is an experience with which most of us are familiar, but a feat few take for granted. Economic prosperity and the appetite of modern society for travel mean that the numbers of flights taken world-wide are increasing by 8% every year. As a result, air travel is fast becoming a major contributor to climate change. The trick of keeping hundreds of tonnes of people and machine in the air and transporting them thousands of miles in an environmentally sensitive way is challenging today's engineers. Through our showcase, we have reached the engineers of tomorrow by introducing an integrated suite of learning activities to young students, designed to inspire, educate and challenge these students in a novel way that we hope will have a lasting impact.
The suite of education tools centres on an exciting 3D presentation of a journey through a jet engine. This novel, stereoscopic journey through an exact virtual replica of the Trent 900 engine explains how an aeroengine works, and introduces some of the fabulous materials and manufacturing techniques which enable the engine to operate at high temperatures. This 3D presentation links with a set of materials-based playing cards and a PC-based engine design challenge where students can design their own jet engine and see if it will take off and fly efficiently.
Developed in collaboration with Rolls-Royce, to inspire children to take up science and engineering, the project has been exhibited at the Royal Society Summer Science Exhibitions in London and Glasgow and was selected for the Queen's 80th Birthday Science Day at Buckingham Palace. Our showcase also attracted interest from EPSRC, winning a grant through its Partnerships for Public Engagement (PPE) competition. Quantitative and qualitative feedback has demonstrated its universal appeal from school children and the general public to practising engine designers, having exhibited to 2284 members of the general public, 2294 VIPs & FRS members, and 3213 school children, teachers and educators. This showcase has stimulated take up of novel elements of the exhibit through a number of channels, ensuring lasting impact and much wider exposure:
The software used to render these images is named the Real-Time Ray Tracer (RTRT). It was developed at the University of Utah’s Scientific Computing and Imaging (SCI) Institute to provide, among other things, a platform for interactively exploring huge data sets. RTRT leads the world in its ability to allow application scientists direct access in interrogating massive data sets. New functionality is continually being added to RTRT (now being folded into a derivative project named Manta); however the core of the original system is in the maintenance phase.
RTRT was designed to run on shared memory architectures. Original development was done using SGI shared memory supercomputers (up to 1024 processors), however with the advent of multi-core processors, it now runs very effectively on commodity hardware. While the movie was rendered at 1280x720 pixels, the ray tracer is usually used at about 600x600 resolution and averages 10 frames per second on a 16 processor (2.4 GHz – Opteron) Linux computer with 64 GB of memory. Smaller datasets can achieve interactivity on a high-end desktop machine.
In the attached video two large data sets are interactively explored. The first data set is of a foam microstructure subjected to compression. The second is a simulation of an explosive device subjected to a transportation fuel fire, and depicts the evolution of the fire, the heating of the steel container and energetic material, the ignition of the explosive and the subsequent rupture of the device. This simulation consists of two main components – a fluid (fire) code that simulates complex chemically reacting flows, and the material (structure) code based on the Material Point Method (MPM). These components are coupled together in C-SAFE’s Uintah computational framework. The foam data set consists of 24 time steps comprising a total of 5.3 GB. The fire/explosion data set consists of 430 time steps for a total of ~60 GB of data.
The ability to visualize and interrogate huge data sets is very important to the scientific community. Visualization is used to verify boundary conditions, look for computational artifacts, and diagnose localized numerical errors. In addition, visualization can be much faster than a more detailed analysis for simple queries such as the velocity of the fluid in a particular region. Using RTRT we can, in real-time, analyze data properties such as shape, characteristic, and fine structure of the simulation with the added ability to explore huge amounts of time varying data.
RTRT allows the volume rendered fire data to be displayed simultaneously with large quantities of particles. In addition, we have utilized ray tracing to predict the visual appearance of a fire using the emission characteristics of the chemical species in the fire and including effects such as scattering due to soot, refraction due thermal variations, and interaction of the consequent light with surrounding objects.
The ability to interact with huge data sets is essential to the success of the modern computational scientist to understand this type of complex data. Mechanisms such as silhouetting, ambient occlusion, and global illumination enhance the ability to perceive subtle shapes that can go unnoticed in common visualization methods. For these reasons, the Real-Time Ray Tracer is a powerful, state-of-the-art tool that may fundamentally change the way scientific data is analyzed in the near future.