Distributed Visualization

How to improve the quality of 3D medical
volume rendering at almost no costs

Steffen Märkle, Axel Spikermann

Dept. of Computer Graphics, FR 3-3,
Technische Universität Berlin,
Germany

INTRODUCTION

One of the actual questions in modern healthcare is how to reduce costs or improve the quality of treatment without extra costs. Today, 3D medical volume rendering is a widely used technique for diagnosis and treatment planning. The quality is given by the used rendering algorithms, image resolution and speed. The presented system improves the quality of medical volume rendering at almost no additional costs.

EXTRA COMPUTATION POWER AT NO COSTS

Generally, to improve visualization systems, extra computation power is needed. Being a general purpose system, any computer can be used for visualization tasks if there is a possibility to put the workload onto it. In normal office environments, the computers are only used part-time and their computation power is fully used only in rare cases. Often there are in the laboratories some older models of workstations that are not used any more. These are resources that can be employed.

The computation task has to be subdivided into small parts that can be computed by one of these additionally employed computers, running a server process. The computers need to be interconnected by a network for the distribution of the tasks.

DISTRIBUTION OF VISUALISATION BASED ON A CLIENT-SERVER-MODEL

Volume visualization is based on the principle of virtually sending rays through a three-dimensional data volume to compute the pixel values of the twodimensional projection. The resulting pixel values can be computed independently for each single ray. Therefore, this process can be easily parallelized.

The presented system is running on PCs, and workstations from different vendors that are connected by 10MBit Ethernet. The distribution is realized by applying the client-server principle: A user starts the (Java-) client program either stand-alone or as an applet in a www-browser. The client connects to one or multiple servers, realized in Java with native ("C"-) libraries for the time critical visualization tasks. Communication is carried out using the IP protocols TCP and UDP.

This highly portable software environment combines the advantages of the object oriented and network-capable Java-system with the execution speed of compiled C-code. Using a Java-applet in the www-browser for the user interface enables world-wide transparent access to the visualization system. The integration into the TUB´s "MediCAL" - System as a visualization tool is also realized by this mechanism.

Several volume rendering algorithms were implemented. Integration of more sophisticated algorithms is actually carried out. Currently, each server receives a copy of the data volume during initialization. Computation of the resulting images is distributed linewise. Each server gets the actual set of parameters for the image generation (viewing transformation, segmentation parameters) and the lines it has to compute. For using computers of different power, the load distribution is adapted dynamically. When a server fails to send its result in a predefined response time, another server is employed.

CONCLUSION

Experimental results show that the overall performance of the presented system for distributed 3D medical volume rendering is increased with each single server added to the configuration. The limiting factor of the system is actually the communcation speed (10 Mbit/s) of our network.

The system uses the hardware being already available at your clinic or department. It is a cheap solution that can provide the practitioner with extra diagnostic quality at almost no extra costs only by taking advantage of existing resources.

As the code is highly portable, almost each computer can be employed as server. Putting the computationally expensive tasks of the visualization onto one or multiple servers, almost any computer that is able to display the images can be used as imaging workstation no matter how fast it is.

Correspondence:

Steffen Märkle
Dept. of Computer Graphics, FR 3-3
Technische Universität Berlin
Franklinstr. 28/29
D-10587 Berlin, Germany

 


Oral presentation at EuroPACS'98, Barcelona, Spain