4. Automated endoscopic capsule analysis using a Grid computing

4. Automated endoscopic capsule analysis using a Grid computing

Automated endoscopic capsule analysis
using a Grid computing environment
Ilídio C Oliveira, Luís Alves, Eduardo Dias, David Pacheco, Sérgio Lima
U. Aveiro & IEETA
João Barros, Miguel P Monteiro, Jorge A Silva
INEB & FEUP
José Maria Fernandes, J P S Cunha, António Sousa Pereira
U.Aveiro & IEETA
http://www.ieeta.pt/sias
Wireless Endoscopic Capsule
(WEC) is emerging in diagnosis
 
6 to 8 hours long video
 
2 hours to review
 
automated image processing
methods can help
 
CADe: polyps, bleeding,…
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
2
Automated Topographic
Segmentation (ATS) can
reduce review times
 
 
Locate the four main
topographic regions
Uses Support Vector Machine
(SVM) classification applied
to MPEG-7 descriptor vectors
► 
 
ATS deployed in CapView
visualization tool
► 
 
http://dx.doi.org/10.1109/
TMI.2007.901430
www.capview.org
Usually takes >1h
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
I
Esogastric
junction
II
Pylorus
III
Ileo-cecal valve
IV
3
The Grid as a “cloud” for capsule R&D
and clinical practice?
 
 
Repositories of
scientific data
► 
GeresMed Project
► 
Data sets anonymized
Processing videos
► 
► 
► 
Run SVM Classification
in Grid infrastructures
Run existing algorithm
“as is”
Domain partition is
feasible
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
4
Enabling friendly Grid
analysis
1. 
Access the MAGI Grid
interfacing portal
2. 
Upload data into MAGI &
select the “CapView” app.
3. 
Data and jobs description
submitted to the Grid
4. 
The progress of jobs is
monitored by MAGI and
output results aggregated
5. 
Classifications calculated
in the Grid used in the
segmentation step
XML
MAGI portal
gLite Services
Ibergrid infrastructure
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
5
Communication between modules
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
6
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
7
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
8
Desktop vs Grid, no paralellization
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
9
No multiplicative factor in the Grid as
the data is partioned
Ibergrid
EGEE
IBM BladeCenter JS21,
with
PowerPC 970@ 2.3
GHz
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
10
High variability in computing times
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
11
Discussion
 
Ibergrid/Ingrid is ready
 
Grid nature includes variability
 
 
► 
High variability in processing times
► 
Significant “overhead” in job preparation
Science requirements are best met
► 
Assyncronous (“fire and forget”)
► 
Massive analysis
Very good speedups with cluster
paralellization
► 
Potential for GPU approaches
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
12
I OLIVEIRA et al | IberGrid2010, Braga - 2010-05-23
13
Automated endoscopic capsule analysis
using a Grid computing environment
Ilídio C Oliveira, Luís Alves, Eduardo Dias, David Pacheco, Sérgio Lima
U. Aveiro & IEETA
João Barros, Miguel P Monteiro, Jorge A Silva
INEB & FEUP
José Maria Fernandes, J P S Cunha, António Sousa Pereira
U.Aveiro & IEETA
http://www.ieeta.pt/sias