Volume Rendering on

footballsyrupΛογισμικό & κατασκευή λογ/κού

1 Δεκ 2013 (πριν από 3 χρόνια και 9 μήνες)

93 εμφανίσεις

A Large Scale Medical
Volume Rendering on
Clustering System

Nopparat Pantsaena, Nont Banditwong, Chuchart Pintavirooj,


Surapan Airphaiboon and Manas Sangworasil

King Mongkut’s Institute of Technology Ladkrabang ,Thailand.

Objective


This paper exploits a concept of the
parallel programming method to
accelerate a medically volume
rendering process by distributing
works to be processed concurrently
on each computer in the clustering
system.

Outline


Introduction


Medical Volume Rendering


Parallel Programming Method and
Clustering System


Experiment and Result


Conclusion

Introduction

Hospital Information System

Medical Image

Viewer

Software

2
D Slice Image Sequences Acquisition System

Result Images from acquisition
system

3
D Medical Image Reconstruction


Reconstruction


Process


2D Cross Section

Slices


3
D Images

Volume Rendering

Volume Rendering Operation

Classification

Gradient Computation

Image
Compositing

Shading

Segmentation

Preparing

Medical Image (CT)

Image Slide Preparing

Segmentation using Histogram

Gradient Computation

Central Difference

Classification

Color and Shading

Q

L

N

Lambert’s cosine Law

I = L

COS
Q

COS
Q
㴠†=
L


N

|I
L
||N|

Image Compositing

Output Image

Parallel Programming Method

and


Clustering System

Clustering System




Clustering system is the technique of
parallel processing that uses several
machines communicating with one
another via high speed network to process
the task.


Clustering system can be implemented
from software or library so
-
called the
parallel programming environments such
as MPI (Massing Passing Interface) and
PVM (Parallel Virtual Machine)

Clustering Category


By Hardware structure


Homogeneous Cluster


Heterogeneous Cluster


By Usage


High Performance Cluster
(Simulation, 3D Rendering)


High
Availability (Web Server,
Machine Controller)

Clustering Architecture and
Parallel Programming
Environment

MPICH

(Message Passing Interface)


We select MPICH to develop the system
because it is a portable programming
running on a wide variety of parallel
computing platforms and it is also a
freeware implementation of MPI standard.


MPICH

It can connect between the
machine in the clustering system.


MPICH

It can pass the

message thought
all process in the system.


Segments The Problem


Load Balancing

Work Pool Scheme

Parallel Algorithm

Load the image slide into
memory

Initialize MPI

Receive the origin of
block

Volume Rendering

Send the output Image to
Manager

Initialize MPI

Pass the origin of each
block

to worker.

Receive output from
Worker

Display the Image

Manager

Worker

Experiment And Result

Experiment And Result



Our heterogeneous system for testing consists
of 6 machines of 2 types


HPx4000
2 Machines.


(CPU Pentium 4 Xeon 1.8 GHz RAM 512 MB
)


HPx2100

4 Machines.


(CPU Pentium 4 1.9 GHz RAM 256 MB)


C
onnected with one another via 100 MBPS
LAN.

Rendering time



0
100
200
300
400
500
600
700
1
2
3
4
5
6
Number of node
Time (sec)
comm
comp
p
T
T
T


.....
(
1
)

System speedup definite of
ideal case.


0
1
2
3
4
5
6
7
1
2
3
4
5
6
Number of node
Speed Up
Speed up
Ideal
p
p
T
T
S
1

…..(2)

System efficiency



0
0
.
2
0
.
4
0
.
6
0
.
8
1
1
.
2
1
2
3
4
5
6
Number of node
Efficiency
Efficiency
Ideal
p
S
E
p
p

…… (3)

Experimental Result

Conclusion



This

paper

focuses

on

parallelizing

the

large
-
scaled

volume

in

volume

rendering

on

clustering

system
.

We

have

shown

the

work

pool

algorithm

for

load

balancing

scheme

ensuring

that

all

processes

are

finished

at

almost

the

same

time
.

This

algorithm

can

attain

high

performance

on

both

homogeneous

and

heterogeneous

systems
.

The

experimental

result

demonstrates

the

efficiency

and

practicality

of

our

algorithm
.

The

clustering

system

can

reduce

time

cost

for

rendering

task
.

The Testing System

Thank you for your attention.