running dicom visualization on the cell (ps3) rsna poster presentation

16
Running DICOM Visualization on the CELL (PS3) Abstract Co-Author: Andre Broekema, BS Presenter: Peter M. van Ooijen, PhD Abstract Co-Author: Matthys Oudkerk, MD, PhD

Upload: broekemaa

Post on 14-May-2015

564 views

Category:

Documents


5 download

TRANSCRIPT

Page 1: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Running DICOM Visualization on the CELL (PS3)

Abstract Co-Author: Andre Broekema, BS Presenter: Peter M. van Ooijen, PhD

Abstract Co-Author: Matthys Oudkerk, MD, PhD

Page 2: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Table of Content

1. The Project (Background)2. Solution in development3. Parts of the solution4. The Project (Evaluation)5. Processed algorithms6. First benchmark7. First benchmark (results)8. First benchmark (evaluation)9. The Project (Discussion)10. Released documents

Page 3: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

The Project (Background)

Visualization is in need of more processing hardware and it is a challenge to realize it with cost effective hardware and free open source software. There are expensive commercial processing software tools available which utilize special computational hardware for accelerating the needed processing power. Aim of this study is to work towards the usage of the CELL Broadband Engine Architecture inside the Playstation3(PS3) and open source software and SDK's to accelerate the visualizations already developed for the free open source community.

Page 4: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Solution in developmentThe solution is designed to be as scalable as possible to allow multiple workstation and Playstations to be added. Client workstations can request processing algorithms on the framework on desired DICOM datasets from a PACS or other DICOM system. Results are then processed through the framework and returned to the client workstations.

Page 5: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Parts of the solution• Workstation

– Linux or Windows based system– GUI for the processing framework– Java client application communication through XML and streaming

• Image server– Linux or Windows based Java service for requesting, receiving DICOM

image datasets from PACS of other DICOM system– Queue system for processing tasks

• Playstation 3– Linux Yellow Dog operating system– Eclipse IDE with remote execution extensions– OFFIS DICOM SDK for reading and writing DICOM files– IBM PS3 SDKs for compiling sources for use on the CELL BE core

Page 6: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

The Project (Evaluation)

By using open source Linux operating systems such as Fedora and Yellow Dog with tools like Eclipse one can create a development platform for use with the PS3. Also the usage of free available SDK's from IBM for developments on the PS3 and the OFFIS DCMTK and the visualization toolkit VTK creates an environment for optimization on the available algorithms inside the toolkits. Currently a framework is running which handles streaming DICOM image data to the PS3, start simple algorithms through dynamically loaded process plug-ins which utilize the DCMTK and stream the result to the requesting workstation.

Page 7: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Processed algorithms

• First processed SUM (Summary) algorithm• Date: 25-03-2009• 75 DICOM Images• 50 Megabytes

Page 8: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Processed algorithms

• MIP (Maximum Intensity Projection) algorithm• Date: 30-03-2009• 75 DICOM Images• 50 Megabytes

Page 9: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Processed algorithms

• Cross section algorithm• Date: 15-04-2009• 75 DICOM Images• 50 Megabytes

Page 10: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

• Cross section algorithm• Date: 28-04-2009• 1000 DICOM Images• 500 Megabytes

Processed algorithms

Page 11: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Processed algorithms

• MIP (Maximum Intensity Projection) algorithm• Date: 28-04-2009• 5000 DICOM Images• 2.5 Gigabytes

Page 12: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

First benchmark• First results of image processing framework on Cell BE without

acceleration cores, only using the Power Processor Unit for calculation.• Date: 19-10-2009

• Dataset 1000 DICOM images (500 Megabytes)• SUM (Summary) algorithm

• Because the Playstation3 has a limited amount of memory to store the DICOM files, a process called “Batching” is implemented. With this process we can split up a large DICOM Dataset with 5000 Images (4,5GigaBytes) in batches of any size, processing plug-ins can request required batches of images to limit the amount of memory needed.

• In this first benchmark we have compared the processing times and loading times of the batch sizes from 5, 10, .. 250 till 500 images against each other without using acceleration cores, only using one Power Processor Unit. (see next slide for results)

Page 13: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

First benchmark (results)

Time durations are in seconds

Page 14: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

• These first results of the image processing framework on Cell BE shows batching images from 1 till 25 creates an overhead in framework activity, from 25 till 50 images shows fastest reading times in this test, when going higher the Operating System requires to swap these images to the Hard Drive when memory is used up.

• The average processing time is 435 seconds. In this 18% is used for file access and 82% used for calculation on the Power Processor Unit. Improvements are expected when using the acceleration Synaptic Processing Units. Results will follow in upcoming benchmarks. Scheduled December 2009.

• Future improvements in file reading times are expected to improve when using a Solid State Drive (SSD). This is currently not a priority with an average of 18% or 50 seconds.

First benchmark (evaluation)

Page 15: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

The Project (Discussion)

Using online available open source processing and visualization toolkits and modify these for performance on cheap processing hardware such as the PS3 is ongoing. When the VTK visualization toolkit is compiled onto the PS3 the next step is to attach it to the process plug-ins to use the available algorithms and move forward to optimize these for use on the CELL architecture. The results will be compared next to other processing hardware and will also be used in developments on other processing intensive projects.

Page 16: Running Dicom Visualization On The Cell (Ps3) Rsna Poster Presentation

Released documents

• http://code.google.com/p/fedora-cell-project/– An Installation Guide for Yellow Dog Linux on the Playstation3– Installing Fedora Core 9 on a i386 X86 PC– Installing the IBM SDK on a Fedora Core 9 PC– Installing the Full System Simulator on a Fedora Core 9 PC– An automated installation script is created and available to install a

development environment for Fedora Core 9 with CELL SDK’s and Eclipse

• http://code.google.com/p/dcmtk-and-eclipse/– Using DCMTK with Eclipse