Viewing the Visible Human using Java and the Web

1
Viewing the Visible Human using Java and the Web

• Paul Coddington
• Advanced Computational Systems Cooperative
  Research Center
• Computer Science Department, University of
  Adelaide
• paulc_at_cs.adelaide.edu.au
• Yuh-Jye Chang (NPAC, Syracuse University)
• Karlie Hutchens (ACSys, University of Adelaide)
• Asia Pacific Web 98
  September 1998

2
The Visible Human

• Digital images of male and female human body.
• Project of the U.S. National Library of Medicine.

• Digitized photos of 1mm axial (horizontal)
  anatomical slices, plus MRI and X-ray CT data.
• Large data set -- raw data over 50 Gbytes, over
  150 Gbytes including reconstructed sagittal
  (side-on) and coronal (front-on) views.
• Male data set available in 1994, female in 1996.

3
The Visible Human and the Internet

• When Visible Man data became available, wanted to
  make the data available via the Internet for
  educational access.
• Main interest was to trial access to large data
  archives from schools with broadband network
  connections, using experimental ATM networks.
• Also want to allow access via existing
  low-bandwidth networks, and even modems.
• Visible Human is a good example of a large
  digital image data archive.
• Data is regular slices, so the problem of finding
  the right image is easier than for a general
  digital image archive.

4
On-line Data Archives

• Many projects working on digital libraries or
  on-line data archives, including OLDA project in
  ACSys CRC.
• Aim to make data sets easily available via
  Internet.
• In just a few years, the Web has become the
  standard interface to on-line data and
  information.
• Problems
• Storing and serving data from large data archives
• Finding the right data (querying and metadata
  standards)
• Interoperability of data archives (interface
  standards)
• On-demand processing of data (active data
  archives)
• Downloading of data (large latencies, limited
  bandwidth)

5
The Visible Human on the Web

• Decided to create a Web interface rather than
  develop specialized client/server application.
• In 1994, only feasible interface was to use
  either HTML forms to enter a slice number, or a
  clickable image map using a picture of the body
  to select the required section.
• Both are slow and clunky interfaces. Everything
  is done by the server, so slow response times.
• CD-ROM versions of Visible Human data provide
  nice interactive graphical interfaces to select
  the image slices.
• Advent of Java enabled development of powerful
  clients for on-line data archives such as Visible
  Human, using interactive GUIs within Web browsers.

6
Client Interface Design

• Intuitive user-friendly graphical interface for
  selecting images.
• Displays thumbnails in 3 view directions for
  rapid browsing.
• Select data set - male or female, photo or MRI or
  CT.
• Move cutting lines to browse and select image
  slices.
• Select resolution for downloading, based on
  available network bandwidth.
• Larger image is displayed in popup window that
  allows panning and zooming.

7
Image Viewer Popup Window
8
Design Issues for Applet Interfaces

• Web is ubiquitous interface, so applet has to
  support wide variety of network bandwidths and
  client computers.
• Must allow choice of image resolution, for speed
  of download and memory limitations on client.
• Interface must be usable on any size monitor, and
  640x480 pixels is not much real estate!
• Must be user-friendly and fairly fast response
  times, or it wont be used.
• Cant access local file system from Java applet,
  but can save image by handing it off to the
  browser or a helper application, and saving it
  from there.

9
Data Pre-processing

• Preprocessing of the image data required the most
  effort in the project (even more than developing
  the Java applet)
• Conversion from raw image format
• Cropping of axial images to reduce size
• Reconstruction of sagittal and coronal images
• Replacing extraneous background material with
  uniform background colour, to improve compression
• JPEG compression to reduce image size
• Multiple resolutions, to support variety of
  network bandwidths
• Thumbnails for the user interface
• Recently processed and added Visible Woman data.

10
Reducing the Data Size

• Crucial to minimize size of the data to be stored
  and downloaded, particularly for current network
  bandwidths.
• Each raw uncompressed axial slice is 7.5 Mbytes.
• Adding sagittal and coronal views increased size
  of data set by factor of 3, to over 150 Gbytes
  uncompressed.
• Standard JPEG compression reduces image size by
  20.
• Lossy JPEG compression OK for educational
  purposes, but medical image archives for
  diagnostic use would need lossless compression
  (factor 2 reduction rather than 20).
• Final data set size is about 3 Gbytes.
• Final image sizes are about 10 - 500 Kbytes.

11
Server-side Processing

• Recently provided capability of on-demand
  server-side processing of the image data, i.e. an
  active data archive.
• Allows cropping of sections of the image (chosen
  using rubber-banding on client), so can quickly
  download small regions (e.g. head) at highest
  resolution.
• Requires uncompressing and cropping of original
  image, then compressing of cropped image for
  transmission.
• Initially done using CGI interface to perl script
  on server.
• JDK 1.1 allows use of Java Remote Method
  Invocation (RMI), but transferring image data as
  object is slow.
• Latest version of applet supports CGI and RMI.

12
Problems with Java

• Write-once, run anywhere not true yet.
  compatibility problems with different JVMs in
  different versions of different browsers on
  different platforms.
• Images are stored in Java as 32-bit integers
  rather than 24-bit RGB or 8-bit greyscale. Can
  cause memory problems on client for large images.
• Dithering of colour images for 8-bit monitors is
  poor.
• RMI is slow, particularly for transferring
  images.
• AWT look-and-feel is different on different
  platforms, but this is now addressed by the Java
  Swing toolkit.

13
Use of the Visible Human Viewer

• Web site has been mirrored by several
  organisations around the world.
• Popular in high schools and medical schools, and
  also with general public.
• Featured in several magazines and newspapers.
• Java source code was made available, and
  variations have appeared which concentrate on
  certain parts of the body, or have additional
  teaching material.
• Has been ported to use Habanero (NCSA) and Tango
  (NPAC), which provide remote collaboration and
  distance learning functionality.

14
Possible Future Work

• Add MRI and CT data (applet interface supports
  it)
• Incorporate anatomical feature database which
  maps features to Visible Human pixels, currently
  being developed by NLM.
• Could provide querying on anatomical features and
  display annotated images of the requested
  feature.
• Add capability to generate animations (MPEGs)
  on-demand for specified regions.
• Add other anatomical data sets, which would
  require more sophisticated server interface and
  metadata support to handle querying on e.g. sex,
  age, medical history, etc.