Å®ÉúСÊÓƵ

DOCTORS and opticians who have no knowledge of computer programming
may soon be able to use techniques of image processing to analyse tumours
and parts of the body, following work by scientists at Canon’s new European
research centre in Guildford.

The researchers have developed a visual programming language, called
VPL, which enables nonspecialists to manipulate an image of the body and
see how they are altering the image from a simple graphic display on a computer
screen.

Until now, doctors who produce an image of a tumour, or opticians looking
at a picture of the retina of the eye, have had to enlist the help of a
specialist in image processing and a computer programmer to enhance their
images and draw out the information they contain. VPL lets them play around
with the image themselves and watch the progress as they vary certain parameters.

This type of language draws on existing programming techniques, but
it is radically different in the way it is presented to the user. It takes
the mystery out of programming by letting the user deal with familiar objects
and concepts, without worrying about the code that is carrying out their
commands. A number of teams of researchers are working on this new generation
of programming languages, which has rarely been applied to complex problems
such as image processing, despite being used to develop video games.

When they process images of the eye, opticians need to compensate for
the blur that the eye’s movements create on the image. The optician produces
a three-dimensional stereoscopic image by combining one two-dimensional
image with another. To do this they have to match the two images subjectively.
This process is far easier if the optician does not have to rely on a programmer
to get it right. ‘Everybody in programming is trying to protect their jobs
by making it more difficult than it really is,’ says David Lau-Kee, the
research scientist who developed VPL.

The visual language was created using Smalltalk, a programming tool
originally developed by Xerox. VPL presents doctors using the program with
a screen display which resembles that of the Apple Macintosh: it uses icons
to represent tasks and displays ‘windows’ on the screen with a specific
function.

The ‘viewer’ function allows doctors to look at an image that they have
previously created – rather like looking at a photocopy of past work.

The ‘display’ function allows the user to pick out a particular part
of an image and display it, for example, on a good colour graphics monitor.

Because the image is simply a series of pixels that vary in intensity,
doctors can adjust an image so that all the parts of it that are brighter
than a chosen threshold value turn black. This creates a binary picture
which makes it easier for the doctor to see some features. Doctors can also
assign false colours to areas of the image which have different values of
brightness.

The screen provides a slider to allow the user to pan between different
settings of these functions and select the best.

The visual language itself deals with a problem in a different way from
conventional programming languages. A ‘functional’ language, for example,
would tackle a problem using a series of mathematical functions. An ‘objectoriented’
language would deal with a problem by creating its own programming analogies
of the real world. It might pull together a description of what a car is,
with information such as the fact that you have to turn the ignition key
to start it.

A visual language would deal with a problem by representing it as a
visual structure, perhaps by simply saying to itself ‘A car looks like this
. . .’.

One of the limitations of the current system is that it can only pick
out those parts of an image which have some form of visual link, such as
having the same brightness. It cannot pick out what looks like a tumour,
on the basis of a medical expert’s hunch. However, Lau-Kee says he could
add this facility if he were to attach some form of ‘expert’ computer system,
containing doctors’ knowledge, onto his basic VPL set up.

A commercial version of VPL could be available in about a year.