![Page 1: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/1.jpg)
Enabling of Grid based Diffusion TensorImaging using a WorkflowImplementation of FSL
Healthgrid 2009, Berlin 30.06.2009
Ralf LützkendorfJohannes BernardingFrank HertelFred ViezensAndreas ThielDagmar Krefting
![Page 2: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/2.jpg)
page 2
Diffusion Tensor Imaging
Diffusion Tensor Imaging• is based on diffusion weighted MRI• allows for tracking of nerve fibers
Main applications• Fundamental research
• Anatomy and formation of axonal tracts• Connectivity between functional areas
• Clinical research• Recognition of desease patterns• Planning support in brain tumor therapy
![Page 3: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/3.jpg)
page 3
Image Processing
Different Software Tools available for DTI
![Page 4: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/4.jpg)
page 4
FDT
FMRIB‘s Diffusion Toolbox:• Graphical user interfaces• Command line tool
Image processing pipeline:• Preprocessing
• Distortion correction• Segmentation
• Local modeling of diffusion parameters (bedpostX)• Estimation of diffusion tensor for each voxel by Markov
Chain Monte Carlo sampling• Computationally expensive• Parallelizable on slice level
• Fiber tractography
![Page 5: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/5.jpg)
page 5
bedpostX
Runtime• Website information: 24 h• Personal computer (512 Mb RAM, 3 GHz, 56 slices): 10 days!• Runtime depends on slicenumber and content!
First approach:• Local windows desktop grid of 10 computers:
• using cygwin• Scales nearly linearly• Runtime ~ 24 hrs• Only temporarily available
Grid approach:• Usage of powerful machines• Parallel processing of MRI slices
![Page 6: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/6.jpg)
page 6
Data Management
Requirements:• Fixed number of dedicated files provided within a directory• Generated by preprocessing• Image Format: NIFTI
SDSC‘s Storage Resource broker:• Linux-like file system structure• Shared access using group rights
Stored data:• Input data• Intermediate results• Final results
![Page 7: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/7.jpg)
page 7
Process Management
Processing steps:• Splitting the data into slices• Run bedpostX• Recombine data
Grid Workflow Execution Service:• Orchestration of the processing steps• Scheduling and resource brokering• Error handling• Reusable workflow templates
![Page 8: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/8.jpg)
page 8
Workflow Layout
Monolithic implementation
Parallelized implementation
![Page 9: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/9.jpg)
page 9
Clinic: User
Workflow Execution Service
Univ. of Leipzig
Data Storage and MetadataBerlin, Dresden, Göttingen
Web-Portal
Uni Leipzig
www.medigrid.de
Grid-NodeZuse Inst. Berlin
Grid-NodeUniv. of Dresden
Grid-NodeUniv. of Leipzig
Grid-NodeGWDG
Göttingen
System architecture
Main components used for application
• All middleware components are free software for academic use
![Page 10: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/10.jpg)
page 10
Results
Runtime measurement:
# Slices Monolithic hrs Parallelized hrs Speedup
11 6.5±0.7 (0.7) 3.5±2.5 (0.3) 1.9
14 14.3±0.8 (0.2) 3.5±1.7 (0.25) 0.9
20 8.9±3.8 (0.4) 3.6±1.2 (0.18) 1.6
56 22.0±8 (0.4) 3.8±2.2 (0.06) 5.6
LocalCluster
56 240.0 (4.2) 23 (0.4) 10.4
Grid
![Page 11: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/11.jpg)
page 11
Results
Reliability:• Complete monolithic grid implementation• Single slice monolithic grid implemetation• Parallelized grid implementation• Five subsequent monolithic grid implementation, tested on
different grid sites• Single slice monolithic local implementation on 64bit machine• Single slice monolithic local implementation on 64bit machine
![Page 12: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/12.jpg)
page 12
Results
Differences between 64bit and 32bit machine
![Page 13: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/13.jpg)
page 13
Results
Accessibility• Webbased up- and download to SRB• Webbased upload, start and control of the workflow• Data and application has been shared between Magdeburg
and Berlin
Usability:• BIRN‘s SRB portlet• GWES workflow portlets• Workflow template (xml-file)• No application specific gui
![Page 14: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/14.jpg)
page 14
Discussion and Outlook
• Speedup by more powerful computing resources• Speedup by parallelization• Fault tolerance essential (poor SRB scalability, network)• Speedup drastically reduced by queuing times• Reliability given for current grid infrastructure• Accessibility and usability sufficient for researcher
Further developments• Completion of the processing on the grid• Provision of a portlet• Usage of more scalable GDMS (iRods ?)• Process optimization (dynamic bundling of slices)
![Page 15: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/15.jpg)
page 15
MedInfoGRID
![Page 16: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/16.jpg)
page 16
Clinic: User
Workflow Execution Service
Univ. of Leipzig
Data Storage and MetadataBerlin, Dresden, Göttingen
Web-Portal
Uni Leipzig
www.medigrid.de
Grid-NodeZuse Inst. Berlin
Grid-NodeUniv. of Dresden
Grid-NodeUniv. of Leipzig
Grid-NodeGWDG
Göttingen
System architecture
User: medical doctor or scientist in medical sciences
• Clinical environment: strict data and network protection
• Not necessarily a computer expert
![Page 17: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/17.jpg)
page 17
Clinic: User
Workflow Execution Service
Univ. of Leipzig
Data Storage and MetadataBerlin, Dresden, Göttingen
Web-Portal
Uni Leipzig
www.medigrid.de
Grid-NodeZuse Inst. Berlin
Grid-NodeUniv. of Dresden
Grid-NodeUniv. of Leipzig
Grid-NodeGWDG
Göttingen
System architecture
Grid portal: gridsphere based webbased access
• Accounts for strict firewall settings (plain http/httpss access)
• Grid access from “everywhere” (with internet connection…)
![Page 18: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/18.jpg)
page 18
Clinic: User
Workflow Execution Service
Univ. of Leipzig
Data Storage and MetadataBerlin, Dresden, Göttingen
Web-Portal
Uni Leipzig
www.medigrid.de
Grid-NodeZuse Inst. Berlin
Grid-NodeUniv. of Dresden
Grid-NodeUniv. of Leipzig
Grid-NodeGWDG
Göttingen
System architecture
Data storage: Storage Resource Broker
• Unix-like access rights (basic data protection)
• Metadata Management
![Page 19: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/19.jpg)
page 19
Clinic: User
Workflow Execution Service
Univ. of Leipzig
Data Storage and MetadataBerlin, Dresden, Göttingen
Web-Portal
Uni Leipzig
www.medigrid.de
Grid-NodeZuse Inst. Berlin
Grid-NodeUniv. of Dresden
Grid-NodeUniv. of Leipzig
Grid-NodeGWDG
Göttingen
System architecture
Grid nodes: Globus Toolkit 4
• Credential based security
• WS-GRAM generic webservice for program execution
![Page 20: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/20.jpg)
page 20
Clinic: User
Workflow Execution Service
Univ. of Leipzig
Data Storage and MetadataBerlin, Dresden, Göttingen
Web-Portal
Uni Leipzig
www.medigrid.de
Grid-NodeZuse Inst. Berlin
Grid-NodeUniv. of Dresden
Grid-NodeUniv. of Leipzig
Grid-NodeGWDG
Göttingen
System architecture
Workflow Manager: Grid Workflow Execution Service (GWES)
• Data driven workflow modeling as high level Petri net
• Resource matching, basic brokering, error handling
![Page 21: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/21.jpg)
page 21
Main tasks for the developer:• Register the software components to the WF-DB• Provide a workflow description (upload or template)
Workflow Manager GWES
Web Services
Grid Workflow User Interface (GWUI)
Grid Portlet
Application Portlet (Genetic Tools, Imaging)
Grid Portlet
GWorkflowDL
Usern
Globus Toolkit 4, Web Services
Run SimpleGlobus Job
WS-GRAM, RFT, SOAP
Web Service
Exist XML DB
<D-GRDL>
D-GRDL
D-GRDL
Run Workflow
Assemble/ MonitorWorkflow
GWorkflowDL
Run Application
MediGRID Workflow Management
<GWorkflowDL>
GWorkflowDL
Grid Workflow Execution Service (GWES)
D-GRDL
Scheduler +Resource Matcher
GRDBDaemon
D-GRDL
MDS, Ganglia + Custom Metrics
![Page 22: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/22.jpg)
page 22
Grid Integration SIESTA DB
Storage of the SIESTA data in the SRB• Preservation of the dedicated folder structure• Group account for shared access• Relevant metadata is stored as
user defined Metadata (Sufmeta)
Integration of the processing tools• Plain command line tools• Execution by invocation of
WS-GRAM automaticallycalled by the GWES
Data transfer• Srb-transfer protocol between SRB and gridnodes• reliable file transfer (RFT) between nodes
ECG ECG presence
ODI Oxygen Desaturation Index
TST Total Sleep Time [min]
AI Apnea Index [/min]
AHI Apnea Hypopnea Index [/min]
Age Age [years]
Sex Sex [male=1,female=2]
Status Health Status [tag]
Height Height [cm]
Weight Weight [kg]
Pulse Pulse rate [/min]
BPsys Blood press. syst. [mmHg]
BPdia Blood press. Diast. [mmHg]
![Page 23: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/23.jpg)
page 23
Processing chain
Sleep-phase resolved heart frequency for study samples, determined by different indicators (age, health status,..)
1. Record selection: Metadata query2. Record analysis
• Heart frequency (HF) analysis on the selected samples• Matching of HF and consensed sleep states• Statistical analysis of individual HF patterns
3. Collection analysis• Statistical analysis on sample set
SRB-Zone
3
GWES
12
![Page 24: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/24.jpg)
page 24
Petri net modeling
Petri net, modelling the application:
Query 2
Query 4
Query 3
RecordSelection
Record Analysis
Record 1Query 1
Query 5
Record 2
Record n
Result 2
Result 1
Result nCollectionAnalysis
Final Result
T1
T2
T2
T2
T3
Transition PlacePlaces & Tokens
![Page 25: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/25.jpg)
page 25
Petri net modeling
Petri net, alternative description
Query 2
Query 4
Query 3
RecordSelection
Query 1
Query 5
Record Result
CollectionAnalysis
Final Result
Record Analysis
n-m m
Control
T3T2T1
![Page 26: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/26.jpg)
page 26
Petri net modeling
Petri net, actual implementation
Problem:Special characters or blanksin token values-> don‘t pass complete
srb queries
![Page 27: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/27.jpg)
page 27
Portal integration
Integration of a workflow template into an applicationspecificportlet:
![Page 28: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/28.jpg)
page 28
Portlet development
MediGRID Portal
![Page 29: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/29.jpg)
page 29
Portlet development
Gridsphere Framework• Gridportlets• Predefined Components• JSR168 compliance
MediGRID Portal• SRB Portlet (from BIRN)• Workflow GUI Portlets
Application specific Portlet:• Workflow list and status• Parameter setting• Workflow initialization• Result visualization
![Page 30: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/30.jpg)
page 30
Application of the tools on heterogeneous longterm measurementsreveals significant flaws in the heart-beat detection
Developer usecase
sqrs sqrs + correction shortterm-FFT
Healthysubjects
Apneapatients
Current usage of the grid: Algorithm testing and development-> parameter scans and evaluation on the datasets
![Page 31: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/31.jpg)
page 31
Discussion and outlook
Clinical ResearcherDeveloper
What are the main findings for this
patient?
Attending Physician
Does my algorithm work for all data?
What are the characteristic patterns
and features?
The application runs on a production grid infrastructure and isusable for all MediGRID participants
A developer can:- test deployed algorithms for different groups/setups (portlet)- perform Parameterscans on the deployed algorithms (workflow)- test algorithms or modifications (deployment to gridnodes)
Main flaws: - Metadata handling -> switching to iRODS- no manual ECG annotation available for SIESTA data
-> creation of ECG annotation-> integration of physiobank databases with annotation
![Page 32: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/32.jpg)
page 32
Discussion and outlook
Clinical ResearcherDeveloper
What are the main findings for this
patient?
Attending Physician
Does my algorithm work for all data?
What are the characteristic patterns
and features?
The application runs on a production grid infrastructure and isusable for all MediGRID participants
A clinical researcher can:- perform statistical analysis on SIESTA data with deployed tools(portlet)- upload and analyse further (pseudonymized) polysomnographies(portlet)
Main flaws: - Portal stability -> switching to LiveRay- rebustness of the algorithms -> developer‘s task- Userfriendly pseudonymization and upload tool
![Page 33: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/33.jpg)
page 33
Discussion and outlook
Clinical ResearcherDeveloper
What are the main findings for this
patient?
Attending Physician
Does my algorithm work for all data?
What are the characteristic patterns
and features?
The application runs on a production grid infrastructure and isusable for all MediGRID participants
A physician can:- upload and analyse (pseudonymized) polysomnographies(portlet)
Main flaws: - Approvement and Validation of the tools -> clinical researcher- Grid access - > simplified user registration- Validation of the infrastructure -> implementation of GCP-rules- Pricing models and accounting -> grid business models
![Page 34: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/34.jpg)
page 34
Thanks to all collaborators
Sebastian Canisius
Univ. Marburg Sleep Center
Andreas Hoheisel
Fraunhofer FIRST
Thomas Tolxdorff
CharitéMedical Inform.
Thomas Penzel
CharitéSleep Center
![Page 35: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/35.jpg)
page 35
Additional slides
Petri Nets
![Page 36: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/36.jpg)
page 36
Mathematical modeling language for distributed systems, consisting of
• Transitions (squares)
• Places (circles), that may hold np tokens (black dots)
• Flow relations (arrows between places and transitions)
• Input place: arrow is pointing from place to transition
• Output place: arrow is pointing from transition to place
• Marking: Distribution of tokens on places
Petri nets
![Page 37: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/37.jpg)
page 37
• Enabling of a transition: • All input places are occupied • All output places may receive further tokens
• Firing of a transition:• One token of each input place is consumed• One token is added to each output place
• Modeling of image processing workflows• Data -> token, executables -> transitions• Program execution -> firing
Petri nets
![Page 38: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/38.jpg)
page 38
Additional slides
Implementation of command line tools
![Page 39: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/39.jpg)
page 39
Implementation steps
Implementation of command-line tools to the grid
1. Deployment of the software to the gridnodes2. Generation of a wrapper script3. Registration of the software4. Creation of a workflow description5. Optional: Integration of the workflow into the user portal
![Page 40: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/40.jpg)
page 40
Software has to be installed on the front-end of the sites• Each application group has it‘s own remote directory
• Copy application from a local directory to the remoteinstallation directory with gsiscp (script)
• Access to the gridnodes via gsissh and svn update
Deployment of software
![Page 41: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/41.jpg)
page 41
A shell-script• Sets environment (pathes, environment variables)• Calls the program(s) • Requirement: all parameters have to be passed as
name/value pair• Program call:
segmentation 51123_1100.png 51123_roi.mat• Script call:
gwes-segmentation-simple.sh–input_image 51123_1100.png –roi 51123_roi.mat
Wrapper script
![Page 42: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/42.jpg)
page 42
Database-entry (exIST-database, dgrdl):• new software (path of the script)• gridnodes where the software is available
D-GRDL Registration
![Page 43: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/43.jpg)
page 43
Xml-based GWorkflowDLgwes-segmentation-simple.sh–input_image 51123_1100.png –roi 51123_roi.mat
Workflow Description
![Page 44: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/44.jpg)
page 44
Additional slides
Further Projects
![Page 45: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/45.jpg)
page 45
Results
Currently implemented:• 5 image- and signalprocessing applications• With application specific portlets:
• Functional MRI: simple workflow (needs matlab)• Virtual vascular surgery: basic interactive visualization• Ultrasound imaging: 4 different workflows• MRE brain segmentation (FSL sienax)
• Without portlets:• Analysis of polysomnographic signals from a clinical study• DTI analysis with FSL
• Prospected: dynamical lung imaging (PneumGRID)
![Page 46: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/46.jpg)
page 46
Additional slides
MediGRID Portal
![Page 47: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/47.jpg)
page 47
Gensequenzanalyse
Grid Certificate
MediGRID User
MediGRID Portal
fMRI
Ontologiezugriff
TRUS
Workflow Management
Hemodynamics
File Browser
Credential Management
SNP Selection
D-GRDL Metadaten-Erstellung D-GRDL Metadaten-
Management
MediGRID Admin
MediGRID Developer
Resource Management
Resource Monitoring
Bioinformatics
Medical Imaging
OntologyAccess
StandardGrid Portlets
Developer-support
Administration
Web Portal
![Page 48: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/48.jpg)
page 48
Additional slides
Requirements of MedicalGrids
![Page 49: Enabling of Grid based Diffusion Tensor Imaging using a ... · • Estimation of diffusion tensor for each voxel by Markov Chain Monte Carlo sampling • Computationally expensive](https://reader034.vdocuments.mx/reader034/viewer/2022042916/5f553590232f1802926d26ee/html5/thumbnails/49.jpg)
page 49
Medical Grids demand special requirements with respect to mere computing Grids
High security and safety
• Patient data, traceability of processing steps
User friendliness
• User accustomed to graphical user interfaces
Virtualization of grid resources
• Heterogeneous data and applications
Current research on modern Grids is working to overcome these barriers
Medical Grids