Medical imaging and visualization toolkit in

Java
Ms Neha Gulati Ms Binul Sharma
Assistant Professor, IT Department BCA IIIrd year
[email protected] [email protected]

platform for researchers and students to

Abstract — Medical imaging research and develop visualization and simulation
clinical applications usually require prototypes, and it also can be used for the
combination and integration of event of clinical applications .
varied techniques starting from image
processing and analysis to realistic
visualization to user-friendly interaction. INTRODUCTION
Researchers with different backgrounds Motivation
coming from diverse areas have been
using numerous types of hardware, The motivation for scripting this paper is primarily an interest
software, and environments to obtain in undertaking a challenging task in a stimulating area of
their results. We also observe that research (JAVA). the chance to find out a few new area of
students often build their tools from computing not covered in lectures.
scratch leading to redundant work. A
generic and versatile medical imaging and
visualization toolkit would be helpful in Problem Definition
medical research and
academic institutes to scale The capabilities of radiology equipment are increasing
back redundant development work and dramatically, and three-dimensional (generally n-dimensional)
hence increase research efficiency. This medical images are getting important and important in clinical
paper presents our experience in diagnosis and therapy. this example drives the research
developing a Medical Imaging and interest in medical image analysis and modeling. This also
Visualization Toolkit (BIL-kit) that's a elevates demands for suitable medical imaging research and
group of comprehensive development platforms and environments which will rapidly
libraries also as variety of interactive turn research ideas into prototypes and clinical applications.
tools. The BIL-kit covers a good range of Medical images are often interpreted as n-dimensional
fundamental functions from image digitized signal sequences. Therefore, a number of the
conversion and transformation, image elemental signal processing algorithms are frequently getting
segmentation, and analysis to geometric used in preprocessing and advanced analysis. due to historical
model generation and manipulation, all reasons and commercial competition also as of researchers’
the high to 3D visualization and personal customs and preferences, images are created,
interactive simulation. The toolkit design communicated, and transferred in numerous formats. the range
and implementation emphasize the of medical image formats has always been an aggravating
reusability and adaptability. BIL-kit is problem in our research experience. During analysis of scans,
implemented within the Java language in various modeling algorithms could also be applied. Finally,
order that it works in hybrid and dynamic the results of study and modeling got to be validated or
research and academic environments. verified by researchers and clinicians, which needs these
This also allows the toolkit to increase its results to be presented during a user friendly and quantitative
usage for the event of Web-based way. to realize this purpose, modeling and visualization
applications. Several BIL-kit-based tools techniques are used frequently to point out the investigation,
and applications are presented including analysis, and modeling leads to the foremost presentable way
image converter, image vital communication bridges between researchers and
processor, anatomy model simulator, clinicians
vascular modeling environment, and
volume viewer. BIL-kit may be a suitable Objective of the Paper
 ● Aims to supply a foundation infrastructure for institutes development is that the Insight Segmentation and Registration
conducting medical imaging researches Toolkit (ITK). ITK is getting to implement the commonly
● Medical image research combined with programing used medical image segmentation and registration algorithms.
language java offers generic image-processing algorithms Similarly, it's developed in C++, and thus, to beat the cross-
starting from simple thresholding, contrast enhancement platform GUI issues, the appliance developers even need to
● The development of a comprehensive toolkit to integrate master a second language like Tcl/Tk, Java, etc. The above-
packages from both areas to supply more comprehensive mentioned systems require a second language or script to beat
toolkits for medical imaging researchers the cross-platform trouble . Moreover, they are not as
convenient as Java when it involves developing Web-based
Objectives applications. There are also medical imaging packages
developed by using Java. Including one is ImageJ—a medical
a. Cover existing technologies for medical imaging imaging software developed by Rasband6 at NIH. ImageJ
b. Brief the proposed technology in java provides a considerable number of basic image operations.
c. Working of BIL-kit with explanations However, most of the image-processing operations are
d. Future and coming approaches actually 2D, sometimes with capabilities to process image
The goal of this work is to deal with these issues to extend stacks. additionally, we found that its 3D modeling and
research efficiency. the stress of toolkit design is targeting the visualization functions aren't sufficient enough to completely
following: support our research activities. Another Java-based medical
1. Portability: The libraries and tools should be platform- imaging package, NeatVision developed by the Vision
independent, in order that the toolkit are often utilized in a Systems Group of Dublin university , provides a powerful
heterogeneous computing environment with minimized visual image-processing flow interface, with strong image-
redevelopment work processing functionality. However, almost like ImageJ, they're
2. Reusability: the planning of the toolkit object classes centralized more on imaging rather than on geometric
should follow the object-oriented principle, in order that the modeling and manipulation.
thing classes are well self-encapsulated , with high cohesion
and minimum coupling.1 Meanwhile, the thing classes shall Java and The Strategy
be tested and documented in order that the users are going to
be ready to use them without knowing much about the Working in a dynamic and fast-pacing biomedical research
implementation details. environment, we have acquired experience to deal with
3. Flexibility: The toolkit shall be a pool of generalized heterogeneous hardware and software platforms and
object classes with the above reusability features, and thus , constructing research prototypes within minimum time limits.
researchers and students can select relevant object classes to Java is designed to meet the challenges of application
quickly assemble a tool or a prototype in their research development in the context of heterogeneous, network-wide
interests. distributed environments. Paramount among these challenges
4. Extensibility: The new modules developed by is secure delivery of applications that consume the minimum
researchers and students shall easily be incorporated into the of system resources, can run on any hardware and software
toolkit.3. View and traverse their list of connections and platform, and can be extended dynamically. Besides the
people made by others within the service. advanced design philosophy of Java language, it has also
brought the quality and productivity to developers. In addition,
RELATED WORK AND OUR APPROACH Java provides some optional packages that are very helpful in
the development of image processing and visualization
Existing Toolkits functions, namely, the Java Advanced Imaging (JAI)
components of Java Media Framework (JMF) and the Java3D
There are several medical imaging toolkits available either component. The advantages of these components will be
within the ownership or the commercial market. One addressed in The Toolkit Architecture and Modules. The
including the commercial packages, the de facto gold Java3D15 component provides the visualization developers
standard in medical image analysis, is Analyze from Mayo. with a set of higher-level APIs compared to most graphics
It's an integrated suite of complementary tools for fully packages which many developers are still using, such as
interactive display, manipulation, and measurement of OpenGL. This feature will significantly reduce the heavy
multidimensional biomedical images. The system is developed burden of the toolkit developers and hence the further
using C++ with Tcl/Tk. VTK3 may be a property right application developers using our toolkit. Network-based
software package developed for 3D graphics and visualization. related medical imaging research and applications are
It's developed in C++ and relies on Tcl/Tk or Java as its cross- increasingly popular. Java, as a language environment
platform graphical user-interface (GUI) support. VTK designed for heterogeneous platforms and distributed
provides only basic image-processing functionality. Besides, computing, is an ideal language in producing network-based
to the developers, they need to master a second language, end-user applications.14 The recent developed technology of
either Tcl/Tk or Java. Another packaged under recent active Java Web Start has delivered a very convenient deployment
solution to deploy Java applications.18 These advanced image acquisition systems. We encountered problems in
features, the seamless compatibility of Java optional dealing with image formats from different vendors in the early
components and the run everywhere promise, allow us to build days; for example, image formats from GE scanners were
a medical imaging and visualization toolkit adapted to different from Siemens. Even different models from the same
dynamic and heterogeneous hardware and software platforms vendor had variations.
of our research environment,. One of the questions debated is This mechanism provides a flexible and useful interface for
Java performance. Indeed, the performance of Java is still not the developers to handle specific medical image formats. It
as good as some optimized languages, such as C++. However, also provides a framework for the additions of format-specific
the speed of Java has been significantly improved in the recent plug-ins. The plug-ins are standard Java classes that are loaded
years and is almost comparable to C++.19 into the API at run time, adding functionality. These plug-ins
In view of software productivity and quality, Java is can also be embedded in Java archive files. Another advantage
observably better than C++. A detailed investigation of Java of the introduced architecture is a powerful API available for
vs. C++ comparison found out Java to be certainly better in handling metadata. The metadata are converted into an XML
both software quality and productivity.20 In addition, a toolkit document. Following is the diagram for java medical imaging
developed by using pure Java requires the developers to technology BIL-Kit Architecture
master only one language; those developers who use other
packages need to know more than one language, such as C++
with another language or scripts, to overcome the cross-
platform problem.

THE TOOLKIT ARCHITECTURE AND MODULES

The Java Medical Imaging and Visualization Toolkit called

BIL-kit developed in our Biomedical Imaging Lab (BIL) has
been built on top of the Java Software Development Kit
(SDK), JAI, and Java3D packages. The toolkit can be divided
into four major modules: foundation module, medical image
input/output (I/O) module, imageprocessing module, and
modeling and visualization module (Figure 2). These modules Fig The Java Medical Imaging and Visualization Toolkit
form the majority of the reusable libraries. On top of these (BIL-kit) architecture
libraries, tools and applications can be built. By using these
tools as templates, and making use of the reusable classes, By extending the Java Image I/O framework, the BIL-kit
more advanced specific applications or prototypes can be built offers support to most of medical image formats, and more
for researchers and clinical users. importantly, it is highly extensible. It can be easily extended
by simply upgrading or adding code (coder/decoder) in the
Foundation Classes Library form of plug-ins to handle new medical image formats in the
future or keep paced with the standards upgrade. This
A comprehensive set of foundation classes has to be built to implementation significantly saves the researchers and
form the base of any toolkit. Java SDK has created a broad developers of doing redundant and tedious work and brings
range of foundation classes that can meet the requirements of great convenience to the users.
medical imaging and visualization in most cases. However, as
BIL-kit is specially designed for medical imaging research and Medical Image Processing and Analysis
applications, some of the frequently used foundation classes Module
needed to be extended from the standard Java SDK, such as
data structures used for presenting medical information and Typical medical image analysis requires various generic
some GUI widgets frequently used in medical imaging. image-processing algorithms; thus, most of them need to be
implemented in the toolkit. However, there are some specific
Medical Image I/O Module requirements to be taken care of. For instance, image
transformation algorithms are frequently required to be
In our research experience, we have encountered numerous performed not just in 2D but also in 3D. Meanwhile, some
image formats from many sources. Besides the general parameters such as pixel size, image acquisition modality, etc.,
photograph formats such as JPG, GIF, and TIFF, the image have to be taken as parameters in image transformation. These
formats used by different computer platforms also vary, such requirements are also considered in the design of the image-
as RGB format for SGI and PNM format for XWindows. processing component in our toolkit. JAI is a set of
Medical images tend to carry patients and diagnosis comprehensive API supplied as an extension package of the
information as well as descriptive information of acquisition standard edition of the Java platform. JAI provides imaging
procedure together. These formats more or less depend on functionality beyond that of the Java Foundation Classes,
although it is compatible with those classes in most All the operations implemented are both in 2D and 3D and are
cases.24,JAI is intended to support image processing using the optimized by our best efforts.
Java programming language as generally as possible. At the
same time, JAI presents a simple programming model that can Modeling and Visualization Module
be readily used in applications without a substantial
mechanical programming overhead or the requirement that the The modeling and visualization component of our toolkit is
programmer has to be expert in all phases of the API’s design. made on top of Java3D component. Java3D may be a 3D
JAI encapsulates image data formats and remote method scene graph supported the graphics programming API for the
invocations within a reusable image data object, allowing an Java language. It is an optional package of the Java standard
image file, a network image object, or a real-time data stream edition. Java3D API provides routines for the creation of 3D
to be processed identically. Thus, JAI represents a simple geometries during a scene graph structure that's independent
programming model while concealing the complexity of the of the underlying hardware implementation for real-time
internal mechanisms. JAI offers several advantages for programming. The API provides scene graph compilation and
application developers compared to other imaging solutions, other optimization techniques. It is heavily optimized toward
e.g., distributed imaging API, which means it is well suited for the wants of real-time 3D rendering. Java3D is an
clientYserver imaging by means of the Java platform’s implementation of the scene graph concepts, which is one
networking architecture and remote execution technologies. It level above elementary special effects packages, like OpenGL.
is also interoperable, which means that it is integrated with the The scene graph is employed extensively in visualization
rest of the JMF APIs, enabling media-rich applications to be development platforms.25 more and more programmers prefer
deployed on the Java platform. However, the JAI package to use the prevailing scene graph programming platforms, like
only provides 2D image operations, which do not utilize full Java3D and Open Inventor, to reduce development efforts and
information contained in 3D medical images. A helpful and costs. Java3D used in our development also uses the scene
practical medical imaging toolkit needs to implement 3D graph. The existing visualization languages and packages
imaging algorithms to take advantage of the latest radiology provide some useful support to create visualization
technology. Image processing in 3D is often considered as a environments for biomedical simulation and education. The
natural extension of 2D image processing, but this is not popular languages and packages used are Java3D, VRML, and
exactly correct. Implementation of 3D image processing needs Open Inventor. These development environments have their
to consider the spatial relation of voxels carefully; sometimes, advantages and disadvantages. We chose Java3D because the
the algorithms need to be redesigned to achieve the optimized foundation of our visualization module due to its several
result.26 As JAI was designed for general-purpose image advantages: making use of the scene graph concept, seamless
processing, it does not support the acquisition of medical integration with Java SDK, and cross platform features.
images. However, as a medical imaging toolkit, it must be Although there are some disputes about the longer term of
able to capture and keep acquisition information, so it can be Java3D, still, Sun Microsystems announced its plan recently to
used in certain automatic image-processing operations. The release the Java3D ASCII text file through some sort of public
initial image-processing algorithms implemented in our toolkit source license within the very near future.30 An expert group
are the following: Y image arithmetic functions, including under the Java Community Process are going to be formed to
addition, subtraction, multiplication and scaling, division, and define and implement new features for the remake of Java3D
blending; Y image algebra transformation, including image API. The open-source strategy may be a good sign that it'll
AND, OR, XOR, INVERT, and bit-shift operations; Y image allow users to contribute bug fixes and utilities, which can
geometrical transformation, namely, rotate or adjust geometric increase the supply and accelerate its pace of development.
size and position of image volumes by resampling 3D image Java3D features a rich set of APIs, which provides powerful
voxels with different interpolation functions; Y spatial domain and flexibly interactive functions useful for building
transformations, including single- and multiple-range educational and simulation systems. It is suitable for building
thresholding, region growing (incremental and adaptive), elegantly 3D models and virtual reality environments.
histogram stretch and histogram equalization, mean and However, to achieve the best results of Java3D rendering and
median filtering, and distance transformation; Y frequency interactive effects, visualization professional knowledge is
domain transformations, including Fourier transformation, required to make use of the Java3D advanced features. To
convolution filtering, which includes a cluster of enhance researchers and students performing on medical imaging and
operations, such as Gaussian and Laplacian filtering, without sufficient visualization knowledge and knowledge ,
sharpening and smoothing, as well as Sobel and zero-crossing certain simplification and abstraction to the Java3D classes
edge detection operations. Y nonlinear spatial transformations, would make their work much easier. Nevertheless, some
such as mathematical morphology including dilation, erosion, comprehensive and abstract objects got to be designed and
opening, and closing. The development of image processing is developed to present multimodal anatomy models. In Java3D,
still in process. We are adding more image-processing the visualization scenario is made as a virtual universe; objects
functionalities. Some more advanced algorithms are on our that impact the visual effect of the virtual universe, like
roadmap of development, including wavelet transformation, geometry, lights, location, orientation, appearance, etc., are
color image processing, and knowledge-based segmentation. formed by nodes and arcs. A node may be a data element, and
an arc may be a relationship between nodes. The scene graph no straightforward data format for storing this data model
is a corporation of the nodes and arcs within the sort of tree because of the complexity of anatomy model presentation and
structure. Nodes are categorized into groups and leafs, where the requirement of catering anatomy information. The more
groups contain transformation and control information, and important thing is the data format defined should be also
leafs contain geometrical information and visual property of flexible and extensible to handle demands from researchers
visual objects. Groups are further subclassified into transform with different research interests and the variety of
group, branch group, shared group, ordered group, and switch applications. We choose XML for this purpose as it is
group. Leafs are further subclassified into shapes that contain designed to describe data in a simple, flexible, and extensible
geometric information, lights that represent light , and lots of way. We have designed an XML schema according to the
others representing the elementary information of anatomy scene graph model described above, which
visualization scenes. The details of Java3D are beyond the incorporates the anatomical terminology along with the XML
scope of this paper and are too detailed for medical image definitions mentioned above in the Image I/O module. This
researchers who aren't visualization professionals. Therefore, schema allows us to store the anatomy models and image data
simplification and comprehensive wrapping to Java3D is in XML files that are understood by the toolkit, so that they
beneficial and necessary to supply tools with friendly and are exchangeable in the tools
convenient interactions for medical background users. In our
toolkit, the scene graph concept is inherited, although it's been BIL-KIT-BASED TOOLS AND APPLICATIONS
enhanced to adapt to the anatomical model presentation We have developed several tools and applications built on the
requirements. We call it the anatomy scene graph. A number toolkit, including DICOM extractor, image converter, image
of comprehensive hierarchical object classes have been processor, general anatomy model simulator, vascular
designed for building the foundation of virtual anatomy modeling environment, segmentation validator, and volume
models for biomedical simulation and education. These are viewer. The present screen snapshots of some of our tools and
enhancement of the Java3D scene graph objects, which give applications built on the toolkit. The image converter
more abstracted and simplified generic medical object models. supporting a wide range of image formats to be decoded and
Some of sophisticated operations and attributes are hidden to encoded. It also provides several useful operations to do
the users to have a friendlier user interface, so that users can simple conversion, e.g., changing data type of pixel
concentrate on medical research rather than to deal with representation or orientation of images. Illustrates the image
visualization techniques. Besides the general geometric processor that can be used for image analysis; many common
elements and surface models, the geometrical shape node in image transformations can be performed in 2D and 3D
the scene graph has been extended to store and interpret more interactively by using this tool. the model simulator where the
complex models, e.g., tubular models and volumetric mesh 3D anatomy scene graph model can be visualized and
models. These models will be very useful in the visualization, manipulated. The user can interactively manipulate the
medical image-based analysis, and simulation applications. An hierarchy relationships of the model objects, their spatial
example of an anatomy model that can be used in visualization positions, and rendering effects. A vascular centerline model
and simulation and its representation of the anatomy scene editing environment where the vascular model can be created
graph. The group node is an extension to the Java3D scene and manipulated. Figure 8 shows an angiographic image
graph transform group node, with some features of switch anatomy labeling tool with the Terminologia Anatomica-based
group and links to an anatomy model incorporated, which index integrated. A 3D cerebrovascular atlas is used to label
allows the visibility of submodels to be manipulated easily. angiography images. A volumetric view combined with a
The shape nodes are elements of anatomy models composed surface-rendered model; the toolkit provides suitable
of the geometry node in Java3D with some functions and functionality for hybrid volume and surface rendering. These
attributes extended for presenting more complex anatomy tools have been developed successfully in a quick manner by
modal data. Integration with Anatomical Knowledge When using BIL-kit, which demonstrates its capacity and potential.
developing applications or prototypes, anatomical knowledge
is often referred, and sometimes, it may also be critically Utilizing the Toolkit
needed, e.g., when building anatomy atlases from various
sources of data. Our toolkit contains the official international Similar to other medical imaging and modeling toolkits, BIL-
guide of anatomical terminology— Terminologia Anatomica kit supports two levels of user interfaces: the developer level
—released by the Federative Committee of Anatomical and the end-user level.
Terminology. Its hierarchy structure is constructed in the
XML format, and a simple built-in database stores the
anatomical names. This feature supports researchers and
developers to label their models with anatomy terminology.
Anatomy Scene Graph Data Schema The anatomy scene graph
model is used for presenting medical anatomy models being
processed or extracted from medical images, so that they can
be rendered and manipulated by the toolkit. Somehow there is
Fig A medical image conversion tool built based on BIL-kit.
Fig An anatomical modeling tool built based on BIL-kit for
manipulating anatomical models
For professionals with software engineering background, they
can make use of the APIs of the reusable Java class libraries,
with their preferred Java development environments,

Fig The anatomy simulator and vascular centerline modeling

environment

manipulate medical imaging data as well as to build 3D

models, visualize them, and interact with them.

Fig A medical image processor built based on BIL-kit. COMPENDIUM AND FUTURE WORK

This paper shares our experience and features our work in

to prototype new algorithms and develop advanced tools and progress on building a medical imaging and visualization
applications. For the users with medical background and no toolkit, which is a useful development and prototyping
programming skills, they can make use of the tools built on platform in our lab. It helps researchers and students to
top of the BIL-kit libraries to conduct their research work much more efficiently by
providing a set of reusable libraries to reduce their redundant
and tedious development efforts. Comparing BIL-kit to the
similar packages mentioned in Related Work and Our
Approach, their scopes of functionality are slightly different;
therefore, it is not easy to make a quantitative comparison.
Our toolkit design has considered the imaging and
visualization requirement from the beginning. This ensures
that the two areas are combined seamlessly, which is an
explicit advantage for developing medical imaging and
modeling tools. This consistent design enriched the reusability
and flexibility compared to other packages. The portability is
achieved by using the Java language.
 modeling are constructed rapidly using an equivalent group of
classes, which demonstrated the toolkit’s reusability and
adaptability . Researchers and students use the reusable object
classes to accelerate their research and development. More
applications to integrate the newest medical imaging and
modeling research results are under development. This
demonstrates its potential as a useful infrastructure in medical
imaging research and its extensibility. the thing oriented
feature of the Java language also plays a crucial part in
promoting its reusability. More work are going to be
performed in implementing advanced image-processing
algorithms, like wavelet transformation and color image
processing, complicated image analysis and recognition
functions, and so on, also as advanced algorithms that are
closely associated with medical domain knowledge, e.g.,
segmentation of anatomical structures. Modeling behaviors
Fig An angiographic image labeling tool with the and properties of anatomical structures also are a possible area
Terminologia Anatomica index integrated to be explored because the toolkit is extensible to support the
info models for physical deformation modeling. Combination
This is much better compared to the packages that required a of the multimodal virtual models with knowledge based and
multiple language combination; similarly, the extensibility is semantic ontology will be an interesting idea to produce
also much better. In view of about 2-year development history genuine and useful virtual models for medical education and
(and a long-term experience in tool development), BIL-kit is simulation. There is still much space to enhance the
still immature. However, the affluent functionality and several functionality to make BIL-kit a fully useful medical imaging
useful tools we developed in such a short period illustrate the and visualization toolkit, acceptable by the research and
benefits contributed by the reusability and flexibility of BIL- clinical communities
kit.

CONCLUSIONS

In this paper, in medical imaging and visualization toolkit in

java we have proposed a novel solution for collaborative
management of shared data. A BIL-Kit was formulated, along
with a multiparty policy specification scheme and
corresponding policy evaluation mechanism and rigorous
programming. In addition, we have introduced an approach for
representing and reasoning about our proposed model. A
proof-of-concept implementation of our solution in progress
called BIL-Kit has been discussed as well, followed by the
usability study and system evaluation of our method.
Fig A volumetric viewer for rendering multimodal models, As part we are planning to examine more comprehensive
including hybrid volume and surface models. Also, we would search more criteria to estimate the features of
our proposed BIL-Kit. Therefore, we would study inference-
the main objective of researchers and students is to conduct based techniques for automatically configure. Besides, we
research on their favorite domain; therefore, requesting them plan to thoroughly integrate the notion of trust and reputation
to find out several different languages isn't cost effective. into BIL-Kit and examine a comprehensive solution to cope
Comparing to those packages requiring various languages to with GUIs and 3D quality for providing top notch medical
require care of either algorithm codes or GUIs on different service in researches and diagnostics.
platforms, the utilization of one language (Java) for both GUI
and algorithms on all platforms may be a more efficient REFERENCES
approach. The advantage of portability is obvious: our tools
are ready to work on various platforms, like MS Windows, [1] Budd T: Understanding Object-Oriented Programming
Linux, Sun OS, Mac OS, et al. , with none modification and with Java. Reading, MA: Addison-Wesley, 1998
even without recompiling. The toolkit is getting used in our
institute from image formats conversion to image [2] Analyze Software, Biomedical Imaging Resource, Mayo
segmentation and analysis, and further to model presentation Foundation:
and visualization. variety of tools for image processing and 3D http://www.mayo.edu/bir/Software/Analyze/Analyze.html,
accessed March 17 2005
[3] Schroeder WJ, Martin KM, Lorensen WE: The design and Visualization Toolkit: An Object-Oriented Approach to 3D
implementation of an object-oriented toolkit for 3d graphics Graphics. Upper Saddle River, NJ: Prentice-Hall Inc., 1998
and visualization. IEEE Visualization ’96, pp 93Y100, 1996
[5] NLM Insight Segmentation and Registration Toolkit,
[4] Schroeder WJ, Martin KM, Lorensen B: The http://www.itk.org/, accessed March 17 2005