Medical Image Storage System For Content-Based Retrieval: Abstract

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Medical Image Storage System for Content-Based Retrieval

Selwyn O. Igwe and Adel S. Elmaghraby


Computer Engineering and Computer Science Department
Speed Scientifc School University of Louisville
Louisville, Kentucky 40292 USA
Abstract- Large image databases have been
designed and used in various applications and in
content-based retrieval systems. Medical Imaging
modalities have been developed in the past each
uniquely designed to specifc imaging modality based
on the image pathobiological medium. This paper
presents an approach to improve the process of image
storage and retrieval based on image data
representation model that integrates both metadata
and content description.
The low level features and the labels associated with
the salient objects extracted fom the mages are
stored in an ORDBMS (Object Relational Database
Management systems) using Oracle. Our technique,
which combines the image descriptors obtained by a
unifed approach phases of image segmentation and
annotation, has a good time complexity and the
experimental results on several image datasets.
Keywords: ORDBS, Database, Medical imaging,
Modality, PLlSQL, Dynamic XLM, HIPPA and
Tablespaces, Metadata, Die OM and Oracle.
I. Introduction
Medical image database for health care
information system has become the bed rock for
quality service and effcient diagnosis of diseases.
The image databases contain detailed patient records,
medical history, clinical data and related diagnostic
images, and post-processed images. However,
advances in computer hardware have made storage of
large collections of digital medical imagery relatively
inexpensive. For design consideration of medical
imaging database system, it is important to note that a
key prerequisite is to have a basic physiological
understanding of the biological medium of the
imaging modality being considered for
storage/archiving. Storage requirements and data
management for a healthcare imaging databases are
huge and enormous. Image segmentation in its self is
a very challenging process. To understand the object
Image, a variety of applications such as object
recognition, image annotation, image retrieval based
on image content, are using the results of a
segmentation method. For retrieval based on image
content, it is essential to have existing image
metadata such as image annotation. For each domain
978-1-4244-9991-5/11/$26.00 2011 IEEE
of imaging application, where image retrieval is
desired, the need for semantic relationship to exist
between the objects present in the images is
imperative.
Content-based retrieval of images is the
ability to retrieve images that are similar to a query
image. Oracle 11g Release 2 visual information
retrieval provides this robust facility. This product is
built on top of Oracle interMedia architecture which
enables storage, retrieval and management of images,
audios and videos. Images are matched using
attributes such as color, texture and structure and
indexing facilitates efcient content-based retrieval
of images by index type. Index type is based on a
multi-level fltering algorithm. The flters reduce the
search space so that the expensive comparison
algorithm operates on a small subset of the data.
Bitmap indexes are used to evaluate the frst flter
resulting in a design which performs well and is
scalable. The Oracle 1lgR2 supports extensible
indexing technology, allowing users to create, use,
and drop instances of this index type as they would
any other standard index. In this paper we present an
overview of the product, the design of the image
index type, and some performance results of our
system.
Figure 1 shows some examples of the
dimensions and size of biomedical Images. Medical
images come fom diferent imaging sources with
diferent physiological and biological features and
sizes [1]. Because of the huge variations in the
images sizes, it is therefore very important for
researchers and clinicians to keep in mind the type of
pathology being investigated for the imaging storage.
This will help in the selection of the appropriate
medical imaging storage feature and modality in
other to efciently and efectively design content
based retrieval systems.
Image
Gray
Ave
Modality Level
Dimensions
(Bits)
Size/xam
Nuclear Medicine 128 x 128
8ar
2MB
16
MRI 256 x 256 12 8 - 20 MB
Ultraound 512 x512 8 5 - 8MB
Doppler
512 x512 24 IS - 24 MB
Ultraound
520
DSA 512 x 512 8 4 -10 MB
CT 512 x 512 12 20 MB
Spiral or Helical
512 x 512 12
40 -150
CT MB
Digital Electronic
Microscopy 512 x 512 8 Varies
(OEM)
Digital Color
Microscopy 512 x 512 24 Varies
(DCM)
Cardiac
512 x 512
500-1000
Catheterization
1024 x 8
MB
1024
Digitized X-ray 2048 x
12 8 MB
Films 2048
Computer 2048 x
12 8-32 MB
Radiology 2048
Digitized 4096 x
12
64 MB (a
Mammogram 4096 pair)
Table 1: Dimensions and size of biomedical Images (Source:
Wong et al 1998)
Medical images such as computed axial
tomography (CT), magnetic resonance imaging
(MRI), digital mammography and positron emission
tomography (PET), all generate a staggering amount
of image data and important medical information. For
example, a single study of CT scan of thirty 512 x
512 x 16 bits images; each examination is
approximately 15 megabytes. And for digital
mammography study of four 4K x 6K x 12 bits, the
amount of data grow to 200 megabytes per
examination [1].
Medical Images and Image related data
(metadata) are fndamental key ingredient in the
medical domain for clinical research, Diagnosis and
pathobiological investigation of diseases. These
images needs to be captured, managed indexed, inter
related and protected. However, as medical image
data grows, the need of a high-performance, flly
automated and highly available storage system is
required to reduce the complexity of the
organization's image and data storage, management
and retrieval. Like wise, many users are also looking
for a storage solution that provides faster response
times and increased data availability and security.
The image storing system will be capable of
storing images in combination of media devices to
achieve a balance between fast delivery overall
storage cost. One of the key elements of any
application design is the system architecture. In the
design of a medical image storage system, one of the
key ingredients is the system architecture piece: It
determines how other interfaces interact and respond
with each other. The Medical Image Archive System
should be a three-tier architecture design.
978-1-4244-9991-5/11/$26.00 2011 IEEE
Storing images in a database is easier to
manage than storing them in the fle system: a simple
benchmark test to see just how much of a difference
it makes in performance was done. Three different
sized thumbnails of an image picture stored in a
directory on my fle system were generated. The
same three images were also stored in an Oracle 109
Express Edition database. Using Firefox to display
the images, the total time it took Firefox to display
each image was measured. Each benchmark was run
10 times for better accuracy.
As shown in Table 2 the images retrieved
fom the database were actually displayed faster than
those retrieved fom the fle system. Speedup fom
this test ranges fom 11 % -15%this benchmark has
been proved to be scalable. The table had only 3 rows
in it. The question is what would happen if you have
1,000 or more images in the database?
Image Size Average Average Speedup
Time (lec.) (lec.)
File System DBMS
Small 6.65 5.90 11.28%
Medium 7.77 6.69 13.90%
Large 7.87 6.68 15.12%
Table 2: Comparing File and Database System
Scenario
There exist some benefts to storing image
and metadata in a database such as: Related
information is automatically synchronized. It is easier
to backup the data when it is all in one place, and it is
also easier to maintain an independent development
environment.
II. The DICOM Standard
The initial goal in developing a standard for
the transmission of digital images was to enable users
to retrieve images and associated information fom
digital imaging equipment in a stadard format that
would be the same across multiple manufacturers. As
we can observe [3], the frst result was the American
College of Radiology (ACR), National Electrical
Manufacturers' Association (NEMA) standard, which
specifed a point-to-point connection. However, the
rapid evolution of computer networking and of
picture archiving and communication systems meant
that this point-to-point standard would be of limited
use. Consequently, a major efort was undertaken to
redesign the ACR-NEMA standard by taking into
account existing standards for networks and current
concepts in the handling of information on such
networks. The Digital Imaging and Communications
in Medicine (DICOM) standard was the result of this
effort. Its popularity has made discussion, if not
521
implementation, of the standard common whenever
digital imaging systems are specifed.
The DICOM standard is extremely
adaptable; a planned feature that has led to the
adoption of DICOM by other specialties that generate
images (e.g. pathology, endoscopy, and dentistry).
The fact that many of the medical imaging equipment
manufacturers ae global corporations has sparked
considerable interational interest in DICOM. The
European standards organization, the Comita
Europaen de Normalization, uses DICOM as the
basis for the flly compatible MEDICOM standard.
In Japan, the Japanese Industry Association
of Radiation Apparatus and the Medical Information
Systems Development Center have adopted the
portions of DICOM that pertain to exchange of
images on removable media and are considering
DICOM for fture versions of the Medical Image
Processing Standard. The DICOM standard is now
being maintained and extended by an interational,
multispecialty committee [2].
The DICOM standard has become the
predominant standard for the communication of
medical images. However, even though the standard
is widely available fom manufacturers and is rapidly
expanding to include non-radiologic imaging, most
radiologists' understanding of it is limited. In part,
this is because DICOM has a "steep leaing curve"
and most introductory material has been written
either for the engineer and is highly technical [2].
Why all the interest in what would seem to
be a simple task? The answer is that it is not as
simple as it frst appears. Most radiologists are
familiar primarily with flm images, and flm can be
viewed anywhere there is a light source. It is the
transition fom flm images to digital images and the
need to communicate, display, and stores these
images that have made DICOM necessary. With flm,
slight differences in exposure, processing, and
viewing will have little efect in these areas. In digital
imaging, however, the diference of a few bytes can
make it impossible to transfer an image fom one
system to another [3].
III. Medical Image Design Considerations
Architectural design has always been the
center piece of any robust computing system.
However, in the domain of imaging system, a number
of new architectural requirements have been
proposed for the imaging architectures of to fture.
In this paper, we review some key requirements for
tomorrow's imaging systems such as: (a) Network
Design Considerations, (b) Storage Device
Considerations, (c) On-line Storage Type
978-1-4244-9991-5/11/$26.00 2011 IEEE
Consideration, (d) System Security Considerations
and (e) Perforance Considerations.
A. Network Design Considerations
The System network should provide
sufcient communication bandwidth that is needed to
support the distribution of multimedia databases, the
graphical user interface for query and retrieval of
database, and the delivery of large medical images
and graphics. A typical Etheret adapter
communicates at 10 Mbits per second (10 Base T).
Various other communication transmission methods
are commonly used today: (Tl-Line), Integrated
Services Digital Network (ISDN), Digital Subscriber
Line (DSL), Frame Relay, and (T3-Line).
Mode Speed Physical Duration
(Bits 1 Sec) Medium
100BASE-T
(twisted pair);
Fast
100 Mbps
100BASE-
40 seconds
Etheret T(twisted pair);
100BASE-
T(optical fber)
Gigabit
1000 Mbps Optical fber 4 seconds
Etheret
cable 512 Kbps
Coaxial cable 19.84 hours
modem to 52 Mbps
ATM 155 Mbps
Twisted-pair
26 seconds
Optical fber
ADSL 2.0 Kbps Twisted-pair
33.33
minutes
BRI: Twisted-
ISDN 128 Kbps
pair
PRJ: T-l or El 8.68 hours
line
II 3.152 M
Twisted-pair, 44.44
coaxial cable minutes
Table 3: Data Transmission Speeds & Mode of Data Transmission
B. Storage Device Considerations
There are a number of storage design
options when choosing the right storage technology
for Medical Image Storage System (MISS). These
storage devices have varying capacities, retrieval
speeds associated to them. For medical imaging, any
storage device must be able to meet the high Security
standard for HlPP A conformation rules and
guidelines, high availability with faster performance,
higher capacities, and greater durability for data
integrity, fexibility and compatibility in complex
computing environment. In picking the right storage
technology, one should consider the continuous and
rapid evolving of storage for medical imaging. Faster
retrieval is expected for all relevant query
comparison in the Picture Archiving and
522
Communication System (PACS) environment.
Depending on the type of digital archive that supports
the P ACS, the retrieval time is measured in seconds,
minutes, or hours. There are several options for long
term storage of digital image data: spinning disks,
magnetic tape, and solid-state fash memory cards
[4]. Which ever design methodology is chosen, the
goal is for a database that maintains consistency and
integrity of data in the database system.
Enterrise storage system architectural
design has impacted by far the fast growing of
networking technology in medical domain. There are
thee existing methods for connecting storage to
computing platforms as shown in fgure 1: Direct
Attached Storage (DAS), Network Attached Storage
(NAS), and Storage Area Networks (SAN).
DAS
Storage
Methodologies
NAS SAN
Figure I: Alterative Storage Methodologies
The intelligent of the storage is dependent
on the sofware. Storage sofware should be able to
perform volume management, fle management,
clustering for high availability, backup and restore,
mirroring ad replication, and application
management. The ability of sofware to make storage
devices more intelligent will lead to devices that are
easier to administer [4].
C. On-Line Storage Type Consideration
RAID is an acronym for Redundant Array of
Inexpensive (or Independent) Disks. A RAID array is
a collection of drives which collectively act as a
single storage system, which can tolerate the failure
of a drive without losing data, and which can operate
independently of each other.
Figure 2: RID Level 5 Model
The RAID Advisory Board has since
substituted "Independent" for "Inexpensive". Since
978-1-4244-9991-5/11/$26.00 2011 IEEE
then, series of papers written by the original three
authors and others has defned and categorized
several data protection and mapping models for disk
arrays. Some of the models described in these papers,
such as mirroring, were known at the time, others
were new. The word levels used by the authors to
diferentiate the models fom each other did actually
suggest that a higher numbered RAID model is better
to a lower numbered one. In the paper, however, I
will be discussing the RAID 5 model since the
prototype database is build on this RAID model.
Figure 2 above, illustrates an example of a
RAID 5 array comprised of three disks - disks A, B
and C. For example, the strip on disk C marked as P
(A 1, B 1) contains the parity for the strips A 1 and B 1.
Also the strip on disk A marked as P (B2, C2)
contains the parity for the strips B2 and C2. RAID 5
technology guarantees that if one of the disks in the
striped set fails, its contents can be extracted using
the information on the remaining fnctioning disks.
Also, a RAID 5 array can improve the thoughput of
read operations by allowing reads to be performed
concurrently on mUltiple disks in the set [5].
D. System Security Consideration
The urgent need to provide and protect the
confdentiality of patient related medical image data
when stored in databases and transmitted over
networks of any kind has been a major key
consideration when designing any medical imaging
system. System security is an essential element in any
medical image system. Here are two ways to provide
confdentiality to a storage or transmission
application. First, confdentiality is based on
mechanisms provided by the underlying
computational infastructure. The advantage is
complete transparency, i.e. the user or a specifc
application does not have to take care about
confdentiality. The obvious disadvantage is that
confdentiality is provided for all applications, no
matter if required or not, and that it is not possible to
exploit specifc properties of certain applications.
Consider the distributed database
infastructure. If the connections among the
components are based on TCP/IP interet
connections (which are not confdential by itself of
course), confdentiality can be provided by creating a
Virtual Private Network (VPN) using IPSec (which
extends the IP protocol by adding confdentiality and
integrity features). In this case, the entire visual data
is encrypted for each transmission which puts a
severe load on the encryption system. The second
possibility is to provide confdentiality is on the
application layer. Here, only applications and
services are secured which have a demand for
523

Strip A4
Strip A3
Parity(B2, C2)
Strip B4
Parity(A3, C3)
Strip B2
Parity(A4, B4)
Strip C3
Strip C2
Strip A1 Strip B1 Parity(A1, B1)
Disk A Disk B Disk C

confdentiality. The disadvantage is that each
application needs to take care for confdentiality by
its own; the advantage is that specifc properties of
certain applications may be exploited to create more
efcient encryption schemes or that encryption is
omitted if not required. Selective encryption of
medical image data takes advantage of the
redundancy in visual data and is therefore classifed
into the second category [6]. The Three basic
components of security as shown in fgure 6 are:
confdentiality, Availability, and Reliability.
Mission-critical databases are ofen large.
To maintain such databases to be available
continuously, the duration to restore any unexpected
failure caused by either hardware or sofware must be
kept to minimum. Sofware that supports snapshot
and multiple backup devices could minimize the
efect on users in case of system failure.
The high availability of any archive system
lies not only in the storage subsystem, disks and
controllers but the host server as well. Server failover
clustering allows one failover cluster to fail over to
any other node in the fail over cluster confguration.
In this way another server can access the data
immediately. This will minimize system down time
and provide high server availability.
I S" I
ConrKlcl til

Dtmclion
Pruction
COllUnption
Figure 3: Security Component in Imaging Storage
E. Performance Considerations
Performance is a critical characteristic of
such a reusable medical database component. A
database component being reused across many
diferent modalities has to satisf the specifc
performance requirements of these modalities in their
deployment environments. For instance, the
performance requirement for image store when
deployed in CT will be diferent fom the
perforance requirement when deployed in
Ultrasound. Similarly the viewing, query and load
performance would vary for each modality.
The key to the performance of any RDBMS
such as Oracle 109 is the underlying database design.
As mentioned in the previous section, Oracle 109
provides effcient object relational mapping for a data
model and generates the database scripts. These
scripts are then deployed on the commercial
978-1-4244-9991-5/11/$26.00 2011 IEEE
RDBMS. An important concept in database design is
normalization. Normalization is the process of
removing redundant data fom the tables in order to
improve storage efciency and data integrity.
Another signifcant parameter for database
performance is the indexing. Queries containing
indexed attributes run more effciently. Oracle 109
provides the option to the user to specif and create
indexes on the various attributes in the data model,
which get refected in the database scripts. In addition
to the performance tuning done at the RDBMS level,
it is also important that the data access layer is
designed for good performance. One of the key high
performance use cases of any medical database is the
fast rendering of a medical image.
Medical images contain pixel data and it is
very important to ensure that this pixel data is stored
for extremely easy and fast access. In our database
component, the pixel data is stored in a separate fle
on the fle system whilst the Meta information is
stored in the database. This gives a dual beneft: Fast
query of the Meta information as the data is stored in
the database, Fast access to pixel data given that the
pixel data is stored on the fle system and fle 10
performance can be maximized when the CPU is
busy processing the metadata information. In addition
to this, the data access layer is also provided with a
fast fle loader in order to maximize the 10 read
throughput of the pixel data. There is also a database
paging algorithm integrated into the data access layer
the retrieves chunks of records fom the database in
an effcient manner without cluttering memory and
blocking the end user for a long time [7].
IV. User Access tool to Medical Image
Storage System (MISS)
Part of my study is based on an attempt to
create an interface to the Digital Database for
diferent imaging modalities for content-based image
retrieval. The user access tool is designed to provide
users access to images in modality of interest. The
user access tool could also be used by researchers and
clinical educators. The goal is to provide a
convenient means by which physicians, researchers
and clinical educators the ability to query a particular
image fom database, view the thumbnail, choose
area of interest and receive image of chosen area in
preferred format. __ __ --'-"" --

.03
+ ~ ~ ~ ~ - + - ~ --~~
. Y
_L
~
Figure 4: User Access Tool for Image Capture and Upload Screen.
524
The user access tool also provides user the
capability to retrieve a whole image, provides users
operations such as to Image registration, browsing
resent searches, and the ability to retrieve and store
some large images in cache. The user access tool also
allows user to manipulate image thumbnails and
areas of interest using their own methods. An
Intermediate database stores thumbnails of images
and areas of interest, and information about the
subject. Access to the database is via the intermediate
database (the middle tier), and image and information
stored in database are retrieved via web form.
Adding Images to the database using the
upload image web page, the page displays a simple
form to collect image description and image location
on the local work station or image source. When the
upload image button is clicked, the browser sends the
image to the web server for storage in the database.
V. Image Database
A prototype Image database system using a
RDBMS Oracle llg Release 2 for Linux Red Hat
residing on Dell Power Edge 840 Server with 2
terabytes of disk storage is used to store the large
binary image objects, metadata and user biographic
profle. The database also stores image thumbnail,
areas of interest and associated imaging text data.
The database supports image querying based on
multiple imaging sets relating to a single subject,
searching and retrieving multiple imaging data sets
based on common features.
The database has Thee Schema based
tablespaces for the following modalities: computed
axial tomography (CT), magnetic resonance imaging
(MRI), digital mammography and positron emission
tomography (PET) and information on images stored
are grouped into textual data which consists of image
metadata, subject's biographic data, the doctors and
the visits with their image, and clinical data are
subject to indexing for fast performance querying and
content-based retrieval. The medical image archive
system will store active (short-term) and static (Iong
term) data on a combination of high and low cost
media in order to achieve a balance between fast
delivery and low overall cost of storage.
For dynamic and fast performance querying,
the database design is based on the following use
case: A subject visits a hospital and completes all
required visiting forms, a triage nurse, then doctors
sees the patients and may or may not send the patient
to the imaging Lab for an images test or sets of
imaging test. To store this event, the information on
the patient, the attending physician and the healthcare
facility will be needed to form this relationship. The
goal here is to build a database structures with a
978-1-4244-9991-5/11/$26.00 2011 IEEE
relationship that is based on low-level database
manipulation language based upon Coddls Relational
Algebra (A set of mathematical operators that
compose, modif, and combine tuples/objects within
diferent relations). Six basic operations; Projection,
Selection, Union, Difference, Product, Rename, Join,
Semijoin Intersection, and Division.
A more detailed view of this physical
database structure with all the attributes and their
types is presented in fgure 5. It is observed that a
subject may or may not necessarily have an image lab
on every visit. For more effcient and robust
querying, the DICOM_IMAGE table has a composite
key made up of three attributes (Image _id, VisiUd
and the modality).
The Medical Image Storage System (MISS)
is a three-tier architecture design. The top tier, the
client layer or the user interface layer, gives a user
access to the application. This layer presents patient
data and images to the user for data manipulation and
data entry. The two main types of user interface for
this layer are the traditional application and the Web
based application. It should be able to provide
versatile data retrieval with a comprehensive
managed index of archival data and a graphical user
interface for seamless data access, manipulation and
display.
HOSPIT
PHYSICi
DlCO:1 L
W WN
*MW.W.D W.
'==.-.
Figure 5: Physical Structure of the Database
The middle tier, the application server layer
or the business logic layer, consists of business and
data rules to assist in resource sharing. Business rules
are executed on the data storage system to manipulate
the database. It should be able to manage the entire
workfow process fom data acquisition to data
repository and to automate the loading, archiving and
subsequent display and delivery of medical images.
The lowest tier, or the Image and data
repository layer, interacts with image and data
usually stored in a database or in permanent storage
medium. The system is able to provide automatic
data management of storage and retrieval (backup,
restore), and immediate and transparent data access.
525
VI. Benchmark Experiments
Image retrieval is an important design
consideration in image storage database applications.
Conventional data management systems are primarily
designed for applications where the data to be
managed is of structured type. In conventional
database management systems, Keywords and
attributes are derived at the time an image is inserted
or uploaded into the database.
To investigate the retrieval abilities of the
(MISS) prototype system, two experimental image
data bases are used. The frst database is a robust
object relational Database Management system
(ORB OMS), containing 600 images of different
modalities such as mammography, ultrasound,
Cardiac (CT), and pathology. Image features and
semantics are represented by using new features such
as set-type attributes, procedural felds, binary large
object boxes, and abstract data types. The second
database, MISS-2, is based on conventional data
management systems architecture, also contains the
same 600 images as in the frst database. In the
second database, the actual images are not managed
by the DBMS, since the latter does not have the
provision for storing the images using database
record felds. The images are stored using the fle
system of the host computer. However, both the
image keywords and attributes are stored in the
database.
This experimental test measures the rate at
which images can be inserted and retrieved between
two different storage scenarios. The two storage
architectures are: an ORDBMS that is based on
extensible and object-oriented approach. Images are
stored in this database based on their semantic and
visual contents. The second database is a
conventional database management systems, were
atomic attributes or keywords are assigned to the
images before insertion into the database.
MISS prototype includes a top-level user
tool that handles both queries and updates to the
database. The querying mechanism retrieves images
on the basis of stored image content as well as on
more complex relations among the stored data. The
RDBMS supports a variety of complex data types
including text, numerical data, and relative, absolute
time, and geographical location. Image retrieval is
based purely on any combination of the complex data
types that are associated with the images, as well as
on the content of the images.
Benchmark test on the prototype database
(MISS) using precision and recall theory was carried
out. The database contains 600 images (fom
diferent modalities) as shown in table 4. Precision
and recall are two performance measures that have
978-1-4244-9991-5/11/$26.00 2011 IEEE
been in use for evaluating content-based retrieval
systems.
Precision as we know is the faction of a
search output that is relevant for a particular query.
Its calculation, hence, requires knowledge of the
relevant and non-relevant hits in the evaluated set of
images. In the context of the present study precision
is defned as the faction of the retrieved Images (the
set A) which is relevant:
Precision = I Rail I AI
Precision = Number of relevant images retrievedl
Total number of images retrieved
Recall on the other hand, is the ability of a
retrieval system to obtain all or most of the relevant
images in the database. Thus it requires knowledge
not just of the relevant and retrieved but also those
not retrieved. The relative recall value is thus defned
as the faction of the relevant Images (the set R)
which has been retrieved:
Recall=IRaIlIRI
Recall = Number of relevant images retrievedl
Total number of relevant images in database
Conte
Key
Modality
# of
nl
wor
Retrieved
Relev Rec
'mag d anI all
Palholo1 80 ? 60 35 43.8
mammo 90 ? II 6 6.7
Ultrsound 90 ? ? 5 4 4.4
Cardiac 50 ? ? 20
16 32.0
CT
Nuclear 50 ? 25
22 44.0
Med
Doppler 70 ? ? 15
7 10.0
Ultra.
Digitized 60 ? 10
6 10.0
Xrav
Radiolo1 100 ? ? 40 30 30.0
DC Micro 55 ? ? 35 15 27.3
Table 4: Precision and Recall measurements of retrieval
performance using content and keyword search
Figure 6: Images with high precision and high recall
measurements showing retrieval performace graph
Precis
-ion
58.3
54.5
80.0
80.0
88.0
46.7
60.0
75.0
42.9
From fgure 6, the query shows results with
higher precision values at lower recall levels and also
a second query result showing higher precision
values at higher recall levels than at lower recall
levels. The query result with higher precision values
at higher recall levels retrieves the image type well,
but does not rank the retrieved images high enough,
526
and does not have a good relevance ranking metric.
The query result of higher precision values at lower
recall levels does not retrieve as well (in terms of
numbers), but rank its retrieved images well, and can
identif good relevance in its ranking metric, but
does not match well to retrieve all or many relevant
images. All queries in fgure 7 are based on content
search instead of keyword search.
m
8

-= ::
Figure 7: Images with high precision and low recall measurements
showing retrieval perfonance graph.
VII. Conclusion
The results show better performance of
MISS image prototype in retrieving images and have
proven to be the best choice based on the benchmark
test. MISS acquired the highest recall and precision
due to the fact that the database system can store and
search on diverse data types, and provide the
fnctionality of an advanced relational database
management system (ARDBMS) such as a high-level
query language, query optimization, and fexible
indexing. Further, the results reveal that structured
queries (i.e. phrased and Boolean) contribute in
achieving better precision and recall. The fndings
also establish the case that precision is inversely
proportional to recall i.e. if precision increases recall
decreases and vice versa. Oracle is used, an object
relational DBMS. As mentioned in above paragraphs,
one of the advantages of using Oracle l1gR in this
research project is the Oracle ARDBMS ability to
create user-defned fnctions and types. These
features are used to perform run-time image analysis
during the querying process.
Experimental results show that storing
images using a Relational Database Management
System (RDBMS), performance is better than storing
them in a fat fle system or traditional databases in
terms of management and processing. The benefts of
this type of design are that: related information on
image is automatically synchronized; it is easier to
backup the data when it is all in one place, it is easier
to maintain an independent development
environment, and additional fle servers can be added
without having to deal with on-the-fy fle
replication.
978-1-4244-9991-5/11/$26.00 2011 IEEE
The classic argument against storing images
in a database is that it is slow. Retrieving images
fom a database is faster than retrieving an image
fom the fle system based on the bench mark results
as described in the previous paragraphs above. This
argument has been proved to be false. The truth is
that images retrieved fom the database via a web
browser will actually display faster than those
retrieved fom the fle system via a web browser.
The user access tool design to database
provides physicians; researchers and clinical
educators the capabilities use the image database
more efciently and productively. Users would be
able to run their scheduled jobs and non-scheduled
jobs. Researches, clinical educators and other
potential users will get access to greater number of
image sample to iterpret for diagnosis and analysis.
The user access tool provides the users the ability to
defne the search rule for their own purpose. The
batch search results will promote the medical image
browse and the dynamic image sharing.
References
[I] Wong, Stephen T. C, Medical Image Databases: Springer
Netherlands, Hardcover - 09-1998, ISBN 0792382897.
Stephen T. C. Wong, Member, lEE & H. K. Huang, senior
Member, lEE: A Hospital Integrated famework for Multi
modality Image Base Management; Systems, Man and
Cyberetics, Part A: Systems and Humas, IEEE
Transactions on Volume: 26, Issue: 4; Digital Object
Identifer: 10.1109/3468. 508824; Publication Year: 1996,
Page(s): 455 - 469
[2] ACR-NEMA Committee. Digital Imaging and
Communications Standard, ACR- NEMA Standards
Publication 300-1985. Washington, DC: National Electrical
Manufacturers Association, 1985.
[3] ACR-NEMA Committee. Digital Imaging and
Communications in Medicine (D1COM): version 3.0.
Washington, DC: National Electrical Manufacturers'
Association, 1993.
[4] Telikepalli R, Drwiega T, Storage area network extension
solutions and their Perfonance assessment.
Communications Magazine, IEEE, 2004, 42 (4), pp. 56-63.
[5] Patterson, D. A, Gibson, G, R. H. Katz, "A Case for
Redundant Array of Inexpensive Disks (RAID)",
Proceedings of the 1988 ACM Conference on Maagement
of Data SIGMOD), Chicago, IL, June 1988, pp. 109-116
[6] T. Maples, G. Spanos, Perfonnance study of a selective
encryption scheme for the security of Networked real-time
video, in: Proceedings of the Fourth Interational Conference
on Computer Communications and Networks (ICCCN'95),
Las Vegas, NV, 1995.
[7] David Power Eugenia, Politou Mark Slaymaker, Steve Harris
Andrew, Simpson Oxford University Computing Laboratory
(UX) 2004, ACM symposium on Applied Computing.
[8] Clarke, S., & Willett, P. (1997), Estimating the recall
perfonance of search engines. ASLIB Proceedings, 49 (7),
184-189.
527

You might also like