CT Brain Perfusion Studies

1. CT Brain Perfusion Studies David Rodríguez González Pedro Santiago del Río Instituto de Física de Cantabria (UC-CSIC) Enrique Marco de Lucas Hospital Universitario Marqués de Valdecilla

2. EGEE ID. Bratislava 19/09/2007 2 Outline Motivation Infarct location Perfusion and hemodynamical maps Int.eu.grid integration Future Work

3. EGEE ID. Bratislava 19/09/2007 3 Introduction Collaboration between IFCA (CSIC-UC) and Hospital Universitario Marqués de Valdecilla at Santander (HUMV) radiologists. Previous collaboration Development of a tool for assisted diagnosis for brain stroke using hemodynamic parameters maps Vendor independent parameters calculation testing several methods Integration in int.eu.grid

4. EGEE ID. Bratislava 19/09/2007 4 Brain Stroke Brain stroke is one of the most important death and disability causes in the EU and USA. Currently available treatments could help to reduce the extension of the problem CT Perfusion techniques can help using the acquired images to generate brain maps showing relevant hemodynamical parameters: MTT Mean Transition Time BV Blood Volume BF Blood Flow But need accurate identification of the ischemic lession Other relevant information for diagnosis is needed: like angiography

5. EGEE ID. Bratislava 19/09/2007 5

6. EGEE ID. Bratislava 19/09/2007 6 STROKE One million strokes occurring per year in the European Union. NINDS trial demonstrated usefulness of thrombolytic treatment of stroke in selected patients rescuing the ischemic penumbra. Thrombolytic treatment has an associated risk of cerebral hemorrhage up to 20%.

7. EGEE ID. Bratislava 19/09/2007 7

8. EGEE ID. Bratislava 19/09/2007 8 Core: part of the ischemic region that is irreversibly injured Penumbra: area of the brain underperfused and in danger of infarcting.

9. EGEE ID. Bratislava 19/09/2007 9 STROKE CT perfusion plays a major role by demonstrating salvageable brain tissue and extension of cerebral core infarction. And… WHAT’S CT PERFUSION??

10. EGEE ID. Bratislava 19/09/2007 10

11. EGEE ID. Bratislava 19/09/2007 11 ANALYSIS Attenuation proportional to blood (= contrast) in brain tissue. Time attenuation curve of reference: -artery (ACA, MCA) -vein

12. EGEE ID. Bratislava 19/09/2007 12 Deconvolution (MTT) Parametric maps CBV = AUC pixel parench/ AUC pixel artery CBF= CBV / MTT

13. EGEE ID. Bratislava 19/09/2007 13 Vascular pathology - Acute ischemic stroke - Chronic ischemia - Vasoespasm Brain tumours MAIN CLINICAL APPLICATIONS

14. EGEE ID. Bratislava 19/09/2007 14 ¿salvageable cerebral tissue??

15. EGEE ID. Bratislava 19/09/2007 15 Infarct location application Using the cerebral blood flow and the mean transit time generate a brain map showing: The infarct core The ischemic penumbra Implemented in Java (ij library for the images) Using input from GE Perfusion application Will use also our maps when finished The program fed with the parameter maps automatically calculates an output map The medic can change the criteria Also can define a good tissue region to use as reference

16. EGEE ID. Bratislava 19/09/2007 16 Infarct location Now a java standalone GUI application It is being ported to the grid as a plugin for the Migrating Desktop Would be linked to the parameter maps application

17. EGEE ID. Bratislava 19/09/2007 17 CT Perfusion For DCE (Dynamic Contrast Enhancement) imaging using CT (aka CT Perfusion) a sequence of images (45) at the same location are taken in a given interval. Concertation Time Curves The objective is to get the three parameters that are related by the central volume principle This is done voxel by voxel

18. EGEE ID. Bratislava 19/09/2007 18 CT Perfusion Application? Input: (4*) 45 CT brain images in DICOM format Using as reference an artery and a vein three parameters are computed for each pixel: blood flow (CBF), blood volume (CBV) and mean transition time (MTT). Prototype in Matlab Using RegTools Toolbox by P.C. Hansen Now implementing the algorithms in ANSI C Java

19. EGEE ID. Bratislava 19/09/2007 19 Deconvolution Problem We want to obtain F and R(t) Ctiss and Cart are the tissue and artery concentrations R(t) is the tissue residue function and it is used to calculate the MTT Numerical deconvolution process is very sensitive to noise in the measured data Inherently ill-conditioned problem

20. EGEE ID. Bratislava 19/09/2007 20 Singular Value Decomposition Discretize the convolution integral equation System of linear equations A is a nxn matriz and x and b are vectors Singular Value Decomposition (SVD)

21. EGEE ID. Bratislava 19/09/2007 21 Regularization Methods TSVD Singular Value Decomposition (SVD) Threshold or truncation index The smaller singular values are eliminated Limits the effects of noise C implementation using SVDLibC Java using JAMA Tikhonov Regularization Prototype in Matlab Java using JAMA

22. EGEE ID. Bratislava 19/09/2007 22 Selection of the regularization parameter Picard plots to estimate Currently using the L-Curve method

23. EGEE ID. Bratislava 19/09/2007 23

24. EGEE ID. Bratislava 19/09/2007 24 Project Vision Distributed Parallel (MPI) Interactive Computing and Storage at the Tera level The main features of this scientific initiative are: - Distributed Parallel (MPI) Interactive Computing and Storage at the Tera level User Friendly Access through a Grid Interactive Desktop with powerful visualization and real simulation steering in real time Supporting Virtual Organizations at all levels: setup, collaborative environment, grid enhancement of applications, execution and monitoring tools. The main features of this scientific initiative are: - Distributed Parallel (MPI) Interactive Computing and Storage at the Tera level User Friendly Access through a Grid Interactive Desktop with powerful visualization and real simulation steering in real time Supporting Virtual Organizations at all levels: setup, collaborative environment, grid enhancement of applications, execution and monitoring tools.

25. EGEE ID. Bratislava 19/09/2007 25 Single sign-on / authorisation Platform independent Batch jobs MPI jobs Running interactive applications using java plugins or VNC Monitoring grid applications Flexible Application framework User profile management Easy application add on Local and grid file management Ideally int.ue.grid will offer to the final user, a researcher, all services through an interface like the migrating desktop, integrating application, infrastructure monitoring if needed, possibility of visualization, etc. It would be fantastic to work through this interface using transparently all the available power to profit and get answer to complex computing tasks in a short time. This would allow to share resources and at the same time attack problems that otherwise could take days. The effort to provide a Migrating Desktop service at a production-quality level is not negligible (Marcin should/could comment). Ideally int.ue.grid will offer to the final user, a researcher, all services through an interface like the migrating desktop, integrating application, infrastructure monitoring if needed, possibility of visualization, etc. It would be fantastic to work through this interface using transparently all the available power to profit and get answer to complex computing tasks in a short time. This would allow to share resources and at the same time attack problems that otherwise could take days. The effort to provide a Migrating Desktop service at a production-quality level is not negligible (Marcin should/could comment).

26. EGEE ID. Bratislava 19/09/2007 26 Interactivity Users can run and steer the application from the Migrating Desktop - a graphical user interface for application management, grid and job monitoring, data and metadata management. Physical parameters and other simulation data can be changed from this graphical interface. Migrating Destop hides the details of most Grid services and allows for setting up and interactively controlling complex systems. This is achieved by means of the Roaming Access Server , which intermediates to the communication with different grid middleware and applications. Next the job in Job Description Language is sent to the Computing Element by the CrossBroker - the component that manages all jobs submitted by users. It supports submission of sequential and parallel jobs. The CrossBroker chooses the most proper Computing Element, where the calculus part of this system is running in the Worker Nodes (WN) of the Grid which fit the resources requirements of the application. Once the application is deployed in those WNs, one of them is designed as Master and the others are the Slaves. The Master is in charge of the communications with all the Slaves and with the MD. The Slaves calculate the particles trajectories, save them on the Storage Element and send them to the Master, which compiles all of them and renders the visualization. The software link between MD and the Master WN over the network channel is Glogin (It is an interactive grid service which offers interactive bidirectional connections between a grid resource and the users desktop machine. This tool communicates both sides in both directions. It captures user events in the MD and forwards them to the Master in one sense and, in the other sense, it outputs the graphical results from the Master in the MD with Java-based GVid video player. It enables Java-based clients to display a videostream which is generated on a remote site and will be used within a plugin for Migrating Desktop. As a result, the grid infrastructure is transparent to the user with very high performances. Users can run and steer the application from the Migrating Desktop - a graphical user interface for application management, grid and job monitoring, data and metadata management. Physical parameters and other simulation data can be changed from this graphical interface. Migrating Destop hides the details of most Grid services and allows for setting up and interactively controlling complex systems. This is achieved by means of the Roaming Access Server , which intermediates to the communication with different grid middleware and applications. Next the job in Job Description Language is sent to the Computing Element by the CrossBroker - the component that manages all jobs submitted by users. It supports submission of sequential and parallel jobs. The CrossBroker chooses the most proper Computing Element, where the calculus part of this system is running in the Worker Nodes (WN) of the Grid which fit the resources requirements of the application. Once the application is deployed in those WNs, one of them is designed as Master and the others are the Slaves. The Master is in charge of the communications with all the Slaves and with the MD. The Slaves calculate the particles trajectories, save them on the Storage Element and send them to the Master, which compiles all of them and renders the visualization. The software link between MD and the Master WN over the network channel is Glogin (It is an interactive grid service which offers interactive bidirectional connections between a grid resource and the users desktop machine. This tool communicates both sides in both directions. It captures user events in the MD and forwards them to the Master in one sense and, in the other sense, it outputs the graphical results from the Master in the MD with Java-based GVid video player. It enables Java-based clients to display a videostream which is generated on a remote site and will be used within a plugin for Migrating Desktop. As a result, the grid infrastructure is transparent to the user with very high performances.

27. EGEE ID. Bratislava 19/09/2007 27 Integration in int.eu.grid The TSVD is already running in int.eu.grid Working on Tikhonov regularization The infarct location Java application would be included in a MD visualization plugin that: Use DICOM images already at int.eu.grid SEs or register new ones (with anonymization) Would run the brain parameters map creation application using TSVD or Tikhonov Using them would create the core and penumbra map.

28. EGEE ID. Bratislava 19/09/2007 28 Other Future Work Implementing alternative techniques for getting the perfusion and hemodynamic parameters: Frequencies filtering Parametrical approaches Bayesian Comparing all these techniques with simulated and real data Using Monte Carlo simulations to check the methods robustness against different levels of SNR Adapting the application for other medical uses of CT perfusion: Brain tumors