The Impact of Visual Programming in Medical Research

as presented at Medicine Meets Virtual Reality II -- Interactive Technology and Healthcare: Visionary Applications for Simulation, Visualization and Robotics, January 27-30, 1994, at the San Diego Marriott Hotel and Marina, sponsored by UCSD School of Medicine

Alan B. Scrivener

abs@well.com

(c) 1993

ABSTRACT


The last five years have seen the emergence of a new  method  for
creating  computer  programs:  visual programming, which augments
and  sometimes  replaces  traditionally   laborious   programming
methods.   Previously  all  programs  were  authored by keyboard-
editing source  code  text  in  a  computer  language.   Now,  by
programming  visually  -- interactively using a mouse to position
and connect software modules displayed on a  graphics  screen  --
application   developers   are   able   to  achieve  productivity
improvements of up to two orders of magnitude, and end-users  are
able to reconfigure and customize programs while they are running
without assistance from the  original  developers.   As  of  this
writing at least four visual programming systems are commercially
available and widely used  on  workstation  class  machines;  the
oldest  and most mature of these is the Application Visual System
(AVS), introduced in  1989.   Medical  researchers  (and  others)
using  this  software tool have reported  that it not only speeds
project  completion,  but  in  many  cases  has   made   possible
additional  projects which would never have been attempted in the
first place without the  productivity  gains  visual  programming
provides.   This  paper describes a number of projects in medical
research which have used AVS, and the ways in which the tool  has
helped them.



WHAT IS VISUAL PROGRAMMING?

A visual programming tool allows a software engineer to  build  a
computer  application  by creating a diagram on-screen, using the
mouse (or other pointing device) to position and "wire  together"
basic  building  blocks,  called  modules, into an interconnected
data flow diagram.  These modules may be little  pieces  of  code
written  in a traditional language (such as FORTRAN or C), or may
be themselves visually programmed from other more primitive  sub-
modules.   The  resulting  on-screen  diagram  then comprises the
visual program.  Literally,  the  picture  is  the  program.  The
creator can execute this visual program at once, interacting with
it as it runs, and can save it to and restore it from a disk file
in  a  human-readable form.   (The saved programs can also be run
in a more traditional non-interactive "batch" mode if needed.)

This new way of programming augments and sometimes  replaces  the
traditional  method of typing characters into a file to produce a
program in some  high-level  language  which  is  first  compiled
(translated into machine language by a compiler program) and then
linked in with other software libraries before it can be executed
(assuming  there  are no syntax errors which prevent the compiler
from successfully completing the translation).

The obvious appeal of visual programming tools is that  they  are
fun  to  use  and  easy to understand.  What is not so obvious is
that  this  new  methodology  has  additional,  sometimes  hidden
benefits.    Rapid   prototyping   becomes  possible,  making  it
practical to do early testing  and  demonstration,  and  to  make
corrections  sooner  in  the  development  cycle  if the software
developer and the end user have  mis-communicated  in  specifying
how  the  program  should  behave.   The  prototype  can later be
further refined to produce the final application -- early work is
not  wasted.   Good  coding  practice  becomes  easier to follow:
modularity is rewarded, and code becomes more reusable, easier to
debug  and easier to document.  Modularity makes debugging easier
, and interactive error checking  --  especially  for  data  type
mismatches -- catches many mistakes immediately, or in some cases
even prevents them. Maintenance is easier, as well as  after-the-
fact user customization, because the visual programs are run-time
reconfigurable.  The pace of work is much faster,  since  changes
made  visually are immediately in effect and testable without the
need for a compilation stage.  Projects which  might  have  taken
weeks or months of traditional programming can often be completed
in hours or even minutes.  (If this sounds like an  exaggeration,
you  should know that I've actually "fudged" this efficiency gain
factor downward  to 100x --  despite  my  own  repeated  personal
experiences  that  it  is  consistently higher -- because it just
sounds so unbelievable.)

Another benefit of visual programs, originally unanticipated,  is
that  they are much easier to adapt to run on parallel processing
systems.   A  visual  program  preserves  information  about  the
"dependencies"  in  an  algorithm.  This is information which the
computer must have available in order to run in parallel,  across
multiple   processors,   without  accumulating  data  errors.   A
traditional program lacks this explicit  dependency  information,
and  the  compiler  must  first  try  to infer and recreate it, a
procedure fraught with  peril,  before  attempting  any  parallel
execution.

In fact, the benefits of visual programming are so  numerous  and
dramatic that one might ask, "Why didn't anybody do this before?"
The answer lies partly in the fact that we have for only about  a
decade   had   available   affordable   personal   computers  and
workstations  equipped  with  necessary   resources:   bit-mapped
graphics  displays,  mice  or  other pointing devices, and enough
speed and memory  to present an complex interactive  visual  user
interface.   But  the  answer  also  lies partly in the fact that
useful breakthroughs are still rare in the area of  human-machine
interface,  and  are  always  obvious  only  in retrospect.  Some
history will help put this in context.


THE EVOLUTION OF HUMAN-COMPUTER INTERFACES

In seeking to understand the somewhat belated development of  the
visual  programming  methodology,  it  is  useful  to look at the
evolution of human- computer interfaces.   (Of  course,  strictly
speaking  engineering  designs  do  not  "evolve"  the  way  that
biological organisms do, but the process of successive refinement
through  an  engineering  cycle  of design and test does resemble
nature's evolutionary cycle of  genetic  shuffling  and  mutation
followed by natural selection.)  Progress in advancing the design
of human-computer interfaces has  proceeded  for  the  most  part
through  periods of very gradual improvement punctuated by sudden
jumps in capability, often triggered by "paradigm shifts" in  the
underlying concepts, models and metaphors.  This halting progress
would've been vastly accelerated if only we had  first  developed
(or  discovered)  what  Warren  McCulloch  called  "a calculus of
intention,"  *   and   Michael   Arbib   called   a   "biological
mathematics." [2]  This would be a notation which mirrors the way
we represent algorithms (computational recipes) in our own minds.
But  today  we  still  lack  such  knowledge  of our own internal
representations.


* This term was passed on to me by Dr.  Gregory  Bateson  at  the
University  of  California at Santa Cruz around 1975; I have been
unable to find it in McCulloch's collected works.  I  did  locate
an  essay on "a logical calculus of the ideas immanent in nervous
activity," in which he asserts: "What these [computing]  machines
lack  by  contrast with the human brain is the ability to operate
in our own natural terms." [1]

We begin instead with  the  original  conceptual  model  for  the
programmable  digital  computer,  described by mathematician Alan
Turing in 1936 [3].  (During World War II  the  British  secretly
built the first implementation of this model, with Turing's help,
and used it in the "Ultra" project for the breaking German codes.
)  His  original paper defined a "Turing machine" as the coupling
of three components: (1) a "finite state machine," which we would
call  the Central Processing Unit (CPU) today, (2) a data storage
"tape" of arbitrary length, corresponding roughly  to  all  of  a
modern  computer's  main  memory,  disk drives, and other digital
storage media, and (3) access to  an  external  "environment"  to
provide  the  machine's  inputs and receive the its outputs, like
today's Input/Output (I/O) sub-systems.

This "environment" for the machine was generally  assumed  to  be
just  another  machine,  either deterministic or selecting from a
set of random behaviors. Turing apparently gave little thought to
the  implications of linking computers and human beings, at least
in his  original  mathematical  formalisms.   In  the  jargon  of
today's  aerospace  engineers who design  "man-rated" systems for
aircraft and spacecraft, the theory of the  Turing  machine  does
not deal with the "man in the loop" case.   (Though -- to be fair
-- later, in Turing's more philosophical writings he did  propose
the  "Turing  test,"  which  provides  an  objective  measure  of
artificial intelligence; when a computer can sufficiently fool  a
person as to its machine identity in a session of limited two-way
interaction it is said to have passed this test.)

Many theorists since Turing also tended to think of the  computer
as  a  system  which  operates  linearly  in  three steps: input,
process and output.  A look at the  Scientific  American  special
issue  on  computers  and  communication  technology for 1966 [4]
finds all of the diagrams having this same structure:

        +-------+    +---------+    +--------+
        | INPUT | -> | PROCESS | -> | OUTPUT |
        +-------+    +---------+    +--------+

Nowhere is the output connected back to  the  input  through  any
explicit  mechanism  at  all,  let  alone  through  something  as
complicated as a human being.

But in practice computer and human were and are  often  connected
in a loop:

        +-------+    +---------+    +--------+
   +--> | INPUT | -> | PROCESS | -> | OUTPUT | >--+
   |    +-------+    +---------+    +--------+    |
   |                                              |
   |                  +-------+                   |
   +----------------< | HUMAN | <-----------------+
                      +-------+

and this loop of human-machine interaction has  brought  its  own
problems.    The   first  of  these  was  the  discovery  of  the
"debugging" process (called simply "pitfalls" at  the  time;  the
actual  name  came  later  when  the  U.S. Navy's legendary Grace
Hopper found a moth stuck in one of a computer's relays.) Maurice
Wilkes of Cambridge was the original programmer of the Electronic
Delay Storage Automatic Calculator  (EDSAC),  the  world's  first
operational  stored-program digital computer.  In his Memoirs [5]
he recalls:

By June 1949 people had begun to realize that it was not so  easy
to  get  the  program  right as had at one time appeared.  I well
remember when this realization came to me with full  force.   The
EDSAC  was located on the top floor of the building and the tape-
punching and editing equipment one floor below...  I  was  trying
to  get working my first non-trivial program...  It was on one of
my journeys between the EDSAC room  and  the  punching  equipment
that "hesitating at the angle of the stairs" the realization came
over me with full force that a good part of the remainder  of  my
life was going to be spent finding errors in my own programs.

Computer historian Martin Cambell-Kelly reports that "no evidence
has  yet  emerged  that  anyone  had  conceived  of the debugging
problem until programs were tried on  real  computers."  [6]   At
Cambridge,  EDSAC  programs  were  initially  debugged by single-
stepping through the code and examining the bits in registers and
memory, a process called "peeping."  A soon as the first debugger
was written for EDSAC by Stanley Grill  in  1950,  "peeping"  was
outlawed  at  Cambridge because it as so inefficient and wasteful
of the machine's valuable time.

When the first computers became  commercially  available  in  the
late 1950s to mid-1960 this sequence of events was recapitulated.
Any commercial institution with a new computer soon found that it
was  very  expensive to operate.  Some system managers complained
that  the  programmers  were  monopolizing  the  computer's  main
control   console   to  do  program  debugging.  Seduced  by  the
responsiveness  of  direct  interaction  with  the  machine,  the
programmers  would  sometimes  thoughtlessly  squander  expensive
computer time, and this frequently prompted the managers  to  ban
programmers  entirely  from  the  computer  room.   They would be
required instead to submit their code  in  hand-written  form  to
non-programming clerks, the first "computer operators," who would
then key in the instructions and run the jobs.  Presumably  these
operators  would feel no temptation to tinker with programs which
they weren't trained to comprehend.  But in fact, this  isolation
strategy often failed when operators started learning programming
on their own.  (This whole syndrome is documented in a collection
of  surviving  corporate  memos from 1965 concerning use of a new
IBM 360, donated by The Traveler's  Group  to  Boston's  Computer
Museum, and seen by this author on display in 1990.)

To this day we still do not  truly  understand  the  dynamics  of
human-computer  interactions.   Nobody  can explain, for example,
why  home  video  game  manufacturer  Nintendo  became  the  most
successful  company in Japan in 1991, supplanting even auto giant
Toyota [7].

The problem with trying to come up with  a  viable  "man  in  the
loop"  theory is that the human is the part of the system we know
the least about.  On the machine side,  a  whole  discipline  has
sprung  up  --  the  theory of computability -- based on Turing's
work.  Consequently the concept of an algorithm is now very  well
understood,   as   are   some  of  the  ultimate  limitations  of
algorithmic   solutions    to    problems.     (Godel's    famous
"Incompleteness  Theorem"  of  1931  [8],  which  proved that any
useful number  system  must  always  include  some  theoretically
undecidable  propositions, is derived directly from the theory of
computability.) But on the human side we are  almost  totally  in
the dark.  The whimsical "Harvard Law" (dealing with lab animals)
states that "under precise conditions of  temperature,  pressure,
humidity  and  all other known variables, the organism will do as
it damn well pleases."  This pretty  much  sums  up  the  current
state  of  our knowledge of our own minds as well.  And without a
reliable theory of human cognition we cannot quantify how  "good"
our  current  human/machine interfaces are; especially we have no
way to measure how "efficient" they are, i.e., how much more room
for  improvement  remains.   (By analogy: if you have gotten part
way to a goal but don't have a  map  showing  you  the  remaining
terrain  --  if  all you know about is your starting position and
your current position Q then you clearly can't tell what per cent
of  the  way  to  the  goal  you  are.  And to further complicate
things, in many situations,  such  as  computer  gaming,  we  are
hard-pressed to define exactly what the goal is to begin with.)

My personal guess is that there is a  great  deal  of  "room  for
improvement"  in the human-machine interface.  Consider that most
types of computer input devices predate the computer itself,  and
many  date  from  the 19th century.  The once- popular punch card
was developed in 1805 for the  control  of  Jacquard  looms.  The
keyboard  goes  back  to the manual typewriter, invented in 1829.
The joystick comes to us from the airplane, an  1890  innovation.
The  buttons  and  paddles  of  today's  video  games  are simply
electrical switches and potentiometers, understood  since  around
the  1820s.   The  audio speaker goes back to the 1876 telephone.
The cathode ray tube dates from 1855, and its use  in  television
is  1930s vintage.  (In fact, so is the NTSC television broadcast
standard which we are still saddled  with  while  government  and
industry  endlessly  debate  High-Definition TV).  Even the high-
tech-sounding "asynchronous serial communications  line"  (as  in
the  still-popular  RS-232 standard) and the near-universal ASCII
code (used  for  storing  and  sending  alphanumeric  text)  both
originally  come  to  us  from  an electro-mechanical device: the
teletypewriter .  First invented in 1914,  the  old  yellow-paper
"teletype"  is  still  fondly  remembered  as  the  workhorse  of
newspaper wire services, as well as for its more recent, somewhat
anachronistic  use (well into the 1970's) as a cheap and reliable
hard-copy terminal for minicomputers. [9]

In fact, I can  only  think  of  a  handful  of  human  interface
peripherals  which  were  designed  expressly  for computers: the
light pen (almost never used any more), the track  ball  and  its
upside-down  counterpart,  the mouse, the touch-sensitive screen,
the magnetic or mechanical 3D space tracker (such as  those  used
in  virtual  reality "data gloves" and "eyephones"), and a rather
obscure knob-like 3D positioning device called  "the  spaceball."
(Even  futuristic  eye-trackers, direct brainwave interfaces, and
electric field muscle motion detectors came first from the analog
electronics  world.)   It  is easy to imagine that there are many
revolutionary  human-machine  interface  devices  waiting  to  be
invented.

The few true computer peripherals we have now are still used most
often  by  the  "computer  operators" of today (now called simply
"the users").  Programmers in the main still  stick  to  the  so-
called QWERTY keyboard interface, named for the first six letters
on the top row, and designed in 1873 to  deliberately  slow  down
typists  who  were  jamming the then-fragile mechanisms.  (Common
letter pairs are on opposite hands, vowels are towards the edges,
etc.)   Most  programming today is done through the bottleneck of
the antique QWERTY keyboard.

And what exactly is being typed?  It is usually  instructions  in
some  programming  language.   The evolution of these programming
languages has been painfully slow at times.  Most of the barriers
have been conceptual.  Amazingly, it took many years in the 1950s
before  the  (in  retrospect)  obvious  innovation  of   assembly
language  programming  was  widely  adopted  This was just a move
away from using pure binary machine language, "raw bits" of  ones
and zeros in base two, to the slight improvement of more readable
letter abbreviations for the  machine  instructions.   Of  course
this  shift was predicated on using some kind of keyboard instead
of a bank of toggle switches to enter programs. *

The initial proposals for a "high level language" were  met  with
skepticism. Programmers asked: who would want to trust a compiler
to write the assembly language code  for  them?   These  skeptics
were  often  caught  up  in  the  mental  challenge of optimizing
programs for the scarce resources of  their  machines  (which  by
today's  standards  were  incredibly slow and had tiny memories).
Assembly language gave them what seemed like total power, tedious
though  it  was to exercise, and they were reluctant to give that
power away.

Finally,  in  1957,  the  first  FORTRAN   (short   for   FORmula
TRANslator)   compiler   was   unveiled   at  Westinghouse-Bettis
Laboratory.  According to computer legend, on 20  April  of  that
year  the  first  user-written  FORTRAN  program was given to the
compiler, and on the first  try  it  returned  an  error  message
complaining  of  a  missing right parenthesis.  The author of the
offending code asked the question: "If the compiler knew what was
wrong,  why  didn't  it  fix  the  problem?"  [10]  That profound
question remains unanswered to this day.


* Surprisingly, the original EDSAC was programmed in an  assembly
language,  even  though other, later computers were not.  But, as
Cambell-Kelly explains: "It was only when a machine  sprang  into
life that the business of programming was seriously considered at
all.  Consequently, users of many early machines  were  obligated
to program in pure hexadecimal [base 16] or some variant.  (It is
interesting that this mistake was repeated in some microprocessor
development systems in the 1970s.)" [6]

In those days, the accepted method  for  writing  a  program  was
first  to manually draw a flow chart, in other words a diagram of
the flow of control,  and then to translate that  diagram  (again
manually)  into  lines of FORTRAN (or other language) code.  Some
of you may remember  when  college  bookstores  sold  flow  chart
drawing  templates  in the computer science section.  This method
was very useful for beginners, but advanced programmers tended to
skip the step of drawing the flow chart on paper; in effect, they
kept it in their heads.  But if you were trying to make sense out
of  someone  else's  code, for instance if you had to maintain or
modify it after they were gone, a copy  of  their  original  flow
chart   was  very  useful.   Therefore  software  managers  began
requiring that their programmers submit  hand-drawn  flow  charts
along   with   completed   program  listings.   Many  programmers
resisted, viewing the drafting of  the  charts  as  "busy  work."
Then  someone  realized  that,  given  a  finished  program,  the
computer could draw the corresponding flow  chart  automatically,
and this seemed to satisfy everyone.

Though it solved short-term procedural and morale  problems,  the
development   of  automatic  flow-charting  obscured  some  basic
issues: it involved  running  the  natural  development  sequence
backwards.    What  was  really needed was a way for the human to
draw the flow chart in a machine-readable way, and then  for  the
computer    to    generate   the   corresponding   program   code
automatically.  I would classify  automatic  flow-charting  as  a
near-miss in the evolution towards visual programming.


THE EMERGENCE OF VISUAL PROGRAMMING TOOLS

Since the 1950s there have certainly been innovations in computer
languages.  Relocatable  object code, virtual memory referencing,
interpreted languages, overlays, multitasking, run-time  loading,
and  object-oriented  programming  leap to mind.  But all of this
has occurred firmly within the  original  QWERTY  paradigm.   The
intervening   decades  have  brought  vast  improvements  in  the
technology for presenting data to the human, especially 2D and 3D
graphics  displays.     But as Ivan Sutherland pointed out in his
landmark 1968 paper, Facilitating the Man-Machine Interface [11]:

The man-machine  interface,  of  course,  has  to  be  a  two-way
interface.   It  turns  out  that  if  you  look  at  the rate of
information flow, the rate of flow from the computer to  the  man
is very high, whereas the rate at which information goes from the
man to the machine is generally much slower...

This fact, largely a consequence of the ubiquity  of  the  QWERTY
keyboard,  has  held  back  the  art of programming.  Even as the
microcomputer  revolution   spawned   the   computer   networking
revolution,  which  enabled the "windows" revolution, programmers
have still created these wonders using the old paradigm.  Despite
a  lot of talk about visual programming throughout the 1980s, few
examples emerged.  I remember two games around 1982 for the Apple
II  personal  computer,  Pinball Construction Set from Electronic
Arts and Sundog from FTL Games, which had something  like  visual
programming  components.  In  that  same time frame a product was
being aggressively marketed for the IBM PC called  The  Last  One
which  purported  to  generate BASIC code from a visually-defined
spec.  (It was supposed to be a software tool so phenomenal  that
it  would  be "The Last One" you ever had to buy.)  It turned out
to be almost total vaporware -- all the demos were faked, and the
company soon vanished. At Rockwell International's Space Division
in 1987 I saw an internal demo of a  primitive  point  and  click
menu-builder  (developed  on  a  VAX minicomputer connected to an
archaic  Tektronix  graphics  terminal);   alas,  it  was   never
completed  or  distributed.   I  still  remember  that when Jaron
Lanier first founded his ill-fated company, he rather impulsively
gave  it  the  name "VPL," for "Visual Programming Language," but
then failed to produce any such products.  By the  late  eighties
Haeberli  [12]  had pioneered the concept of a data flow network,
and demonstrated a "proof of  concept"  in  the  program  ConMan,
which had limited capabilities and availability .

The first commercially available visual  programming  system  (to
this author's knowledge) was the Application Visualization System
(AVS) version 2, in 1990. Initially AVS version 1 was released in
1989, by Stellar Computer Inc., as a point and click 3D geometric
rendering program for its high-performance workstations [13].  It
lacked  any  visual programming capability.   A year later a team
of  mostly  the  same  developers  produced  VEX,   the   "Volume
Exploratorium,"   with  a  true visual programming interface, and
they demonstrated it at  the  Chapel  Hill  Volume  Visualization
Workshop in 1989 [14]. They then grafted VEX onto original AVS to
produce  a  visually-programmable   3D   rendering   and   volume
visualization system.  This was released by Stellar as AVS2 .  In
the intervening three  years  there  have  been  four  additional
software  releases, bringing us to the current version, AVS 5.01,
which shipped in December of 1993.  Also during this time the AVS
program  has  been  ported  to  every major brand of workstation,
becoming the most successful product Stellar ever  produced,  and
as of the beginning of 1992 a separate software company, Advanced
Visual Systems (AVS) Inc., was "spun  off,"  with  a  charter  to
continue to develop and evolve visual programming tools.

Since the original  release  of  AVS2  several  competitors  have
appeared  on the market , including Khoros from the University of
New Mexico,  IRIS Explorer from Silicon Graphics Inc. (SGI),  and
Data Explorer (DX) from IBM Corp. Principal architects of most of
these programs shared the stage at a SIGGRAPH92 panel  on  visual
programming.   Some analysts have predicted that IBM's entry into
this business will  "legitimize  the  market."   Perhaps  a  more
credible  "legitimization"  occurred when  the U.S. Environmental
Protection Agency (EPA),  the  United  States  Geological  Survey
(USGS), and Sandia National Labs of the U.S. Department of Energy
(DOE), all adopted AVS as their standard scientific data analysis
and visualization tool.



USES OF AVS IN MEDICAL RESEARCH

But the focus of this conference, and this talk,  is  on  medical
research.   I'd  like  to  take  you  on  a  quick  tour of a few
institutions where AVS is being used to speed the development  of
the  software  components of projects exploring various fronts on
the leading edge of medical research.  (I am familiar with  these
particular  examples  because  of  my  involvement  with  the AVS
product for the last five years as a field technical expert.   Of
course, none of this research is my own; this represents a survey
of what others are doing, based on what  they've  published,  and
they things they've told me personally that I have been permitted
to share.)  Most of these groups have two things in common:  they
don't  have  a  lot of people available (especially programmers),
and they're in a big hurry. (After all, human suffering continues
to  be  a powerful motivator.)  All of them have embraced AVS for
its  ability  to  empower  the  researchers  to  achieve  greater
software development results with fewer resources.


REHEARSING RADIATION TREATMENTS AT THE UNIVERSITY OF MICHIGAN

Researchers at the University of Michigan at Ann Arbor have  been
developing treatment planning systems for radiation oncology beam
therapy for over five years now.  Their  starting  point  was  an
existing 3D treatment planning system, U-Mplan, which used CT and
MR data to visualize a tumor in  its  geometric  relationship  to
planned  radiation  beam  positions;  though  it  had some of the
functionality they wanted (including  robust  dosage  calculation
algortihms) it lacked the kind of 3D displays they wanted, and it
was hardware-dependent and had a monolithic, unextesible software
architecture.   They  wanted  to enhance the capabilities of this
system to include better interactive controls,  including  manual
outlining  of  tumor  boundaries  on  2D  slices,  and ability to
manipulate a 3D view of the resulting treatment  plan  --  either
from  an  omniscient,  or a beam's-eye view.  They also needed to
combine displays of 2D grey-scale images and full 3D surface  and
volume  display  methods,  as well as generating traditional line
graphs.  Lastly, they wanted the resulting system  to  run  on  a
wide  variety  of  hardware  platforms,  to  avoid  the  hardware
obsolescence problems that have plagued medical software  systems
in  the  past.   By basing their development efforts on AVS, they
were able to achieve these goals, quickly prototype and test  the
system,  and  concentrate  their  efforts  mainly  on the medical
research  side,  instead  of  working   on   extensive   software
infrastructure.   The resulting system offers an improved ability
to  concentrate  radiation  mostly  at  the  tumor  site,   while
generally  sparing  surrounding tissue, and also allows treatment
plans to be produced and verified much more quickly. [15]



FINDING THE SHAPE OF A DEADLY VIRUS AT THE WISTAR INSTITUTE

Researchers  at  the  Wistar  Institute   in   Philadelphia,   in
collaboration   with   others   at   European  Molecular  Biology
Laboratory  (EMBL)  in  Heidelberg,  Germany,  were  seeking   to
understand  the  structure  of the adenovirus, which is a primary
cause of death in third world children.  Most children who die of
"starvation"  actually  perish  from  the  dehydration  caused by
diarrhea from adenovirus infection.  Due its size and complexity,
the   structure  of  the  virus  could  not  be  determined  from
crystallography techniques which had worked on  simpler  viruses.
Using a novel approach, the researchers took electron micrographs
of the viruses in sections, and then used  AVS  to  combine  them
into  a  3D volume for visualization.  This procedure yielded the
first pictures of the virus as  a  solid  object.    Knowing  the
molecular structure of some of its proteins, the researchers have
begun to reconstruct  its  complete  molecular  structure.   This
research  may  someday  lead  to a life-saving vaccine.  The team
also believes the techniques they have  developed  may  apply  to
cancer  and  AIDS viruses.  The use of AVS in the project allowed
them to concentrate on the problem, instead of  the  intermediate
software  development  requirements  which might have overwhelmed
them if  they'd  used  more  traditional  non-visual  programming
methods.  [16, 17]


HOLOGRAMS OF THE INSIDE OF THE BODY AT VOXEL CORP.

Voxel Corporation has developed the first true laser  holographic
display   of   medical  volume  datasets  (CT  and  MRI).   These
"voxgrams" show the entire volume at once, instead of  2D  slices
which  show  only  a  tiny  portion  of  a  volume,  or 3D volume
renderings which may obscure features not in front.  AVS was used
to  quickly  develop the interactive user interface needed in the
preparation of the data for output  to  the  holographic  camera.
[18]


PICTURING PROTEINS IN STEREO AT THE SCRIPPS RESEARCH INSTITUTE

Researchers at  the  Scripps  Research  Institute  in  La  Jolla,
California  (the  largest private research institute in the U.S.)
have been studying a variety of large biological molecules at the
institute's  Molecular Biology Laboratory, under the direction of
founder Arthur Olson.  Part of their mission has been  to  expand
the  use  of  computer  graphics in molecular structure research.
[19]  They have been using AVS in this mission since 1988, and it
has  aided  in  understanding the immune system, the processes of
antibody docking, and improving procedures for drug design.  [20]
One  of  their  pet projects is the use of 3D stereo displays for
showing molecular structure.  They have also pioneered the use of
a  graphics  technique  called  "texture  mapping"  for  labeling
protein residues in molecules.

The rapidity with which the can customize AVS for their uses  was
dramatized  recently  when they were asked by Scientific American
to  produce  a  cover  illustration  of  a  radical   interceptor
antioxidant  molecule  to  accompany  a  survey  article  on  the
chemical basis of aging, and they  were  able  to  respond  short
order.  [21]


DAY-GLO CELLS IN A 3D MICROSCOPE, RECOGNIZING MONKEY BRAINS,  AND
MAPPING THE CHAOS IN A HEARTBEAT AT THE UNIVERSITY OF CALIFORNIA,
LOS ANGELES

Many separate workgroups at UCLA are  using  AVS;  some  of  them
since  it  was  first  released.   Researchers  there  can buy it
through a specially-priced campus program,  and  today  the  UCLA
campus  represents  the  largest  installed   base  of  AVS users
outside of U.S. government labs.    Of  the  eighteen  (or  more)
software  licenses  on  campus,  at  least  three copies in three
different departments are devoted exclusively to medical research
.

A team of researchers lead by Andy Jacobson is exploring confocal
microscopy  with  a  patented technique using fluorescent dyes to
highlight features of  cell  metabolism.   The  results  of  this
research  are  not  yet published, but follow in the footsteps of
some prior work which  cited  in  the  notes.   [22]   Meanwhile,
Jacobson  has  been  experimenting  with  the volume rendering of
confocal data,  and  sharing  the  resulting  images  with  other
researchers at UCLA.

Kelby Chan of the Department of Radiology has been  experimenting
with  the  feasibility  of  fully digital 2D radiographs for some
time using AVS.  More recently, he has developed a new method for
semi-automatic  tissue  segmentation of volume data, using neural
net methods.  [23]  Segmentation is the  process  of  identifying
which  portion of a volume of data corresponds to what anatomical
features, and has proved  to  be  a  very  difficult  problem  to
automate.  This  new  method  involves  a clinician identifying a
point which is within a certain structure, and the software  then
guessing  which  other  points  in  a volume are part of the same
structure.  Early tests of the method have been done  on  3D  MRI
scans  of  monkey  brains,  segmenting  out the brain tissue from
skull, skin, and other  soft  tissues.   AVS  has  permitted  the
research  to  be  focused  on  the development of new algorithms,
while the problems with more mundane software integration  issues
--  including  migrating  to  a  new  hardware  platform part way
through the project -- were kept to a minimum.

Dr. Alan Garfinkel studies applications  of  mathematical  tools,
especially   chaos  theory,  to  biological  systems.   He  first
acquired the AVS software for use in 3D volume  rendering  of  CT
and  MRI  scans  as well as confocal microscope data. But once he
saw the power of the tool, he began to use for other projects  as
well.  He has been hampered by the fact that for the last several
years, every time  he's  got  an  undergraduate  trained  in  the
programming   and   operation   of   his  computer  systems  that
undergraduate has been accepted into a medical school in  another
city  and  Dr.  Garfinkel  has  had  to  start  over with another
programmer. (This cycle has continued yearly as long  as  I  have
known   him.)   He  says  that  the  comparative  simplicity  and
modularity of visual programming in AVS has made it possible  for
him to bring each new programmer up to speed in time to make much
new forward progress before the next iteration occurs.

One of the projects he applied AVS to was  a  kineseology  study,
comparing  the gaits (patterns of walking) in normal and impaired
persons.  Using AVS to visualize joint angles  in  a  phase-space
representation,  he  found  that  an impaired walker's gait has a
lower dimension (in a fractal sense).  This adds evidence to  the
emerging  generalization that healthy biological systems are more
chaotic than sick ones.  (Unfortunately, this  research  has  not
yet been published.)

More recently, Dr. Garfinkel has collaborated with others in  the
first  demonstration  of  real-time  control  of  cardiac  tissue
fibrillation, using the mathematics of chaos theory.   [24]   AVS
was  used  to  make  phase  portraits  of  the  data  from  these
experiments, using techniques  similar  to  those  developed  for
displaying  the  gait  data.   One  of  these phase portraits was
featured on the cover of Science Weekly  in  association  with  a
survey   article  on  controlling  biological  chaos.   [25]  The
research may eventually make possible the  design  of  much  more
effective and lower power-consuming cardiac pacemakers.


CAPTURING CHAOTIC TREMORS WITH A VR DATAGLOVE, SEEING THE  SHAPES
OF  BRAINWAVES  IN SPACE-TIME, AND THE THERAPEUTIC USE OF DIGITAL
BUTT-PRINTS AT LOMA LINDA UNIVERSITY MEDICAL CENTER

With a grant to study therapies  for  patients  with  Parkinson's
disease,  researchers at the Department of Neurology (since moved
to Rehabilitation  Therapy)  at  Loma  Linda  University  Medical
Center  first  obtained  AVS  for analyzing electroencephalograph
(EEG) data ("brainwaves") taken from these  patients.   Initially
they  used AVS to develop and refine algorithms for measuring the
auto-correlation of these recorded signals, phase-plotting a data
steam  against  a  time-lagged  copy  of  itself  to give a rough
approximation of the "fractal dimension" of the data.  Confirming
the  results  of  Chinese investigators [26], the Loma Linda team
found that data recorded from  a  Parkinson's  patient  during  a
tremor  has  a  lower fractal dimension than data taken while the
patient is in a more normal neurological state.

Meanwhile,  through  aggressive  networking  and  self-promotion,
these  researchers  convinced  VPL  Corp.  to donate one of their
second-generation DataGlove devices to the lab.  This novel input
peripheral,   developed   for   use   in   Virtual  Reality  (VR)
applications, provides three dimensions of hand  position  (X,  Y
and  Z),  and  hand orientation (roll, pitch and yaw), as well as
the flexion and extension angles of the ten major finger  joints,
all delivered to the computer as an incessant stream of numerical
data up to 60 times per second.  Hooking the device up  to  their
Parkinson's  patients,  the  team  became  the  first  to use the
DataGlove  as  a  true  medical  instrument  which  records   raw
physiometric  data, instead of for its primary design function as
a  purely  volitional  mechanism  for  interacting  with   a   VR
application through deliberate pointing and gesturing.

Once the hand motion data was acquired from  a  patient,  it  was
easy  for the investigators to use their existing visual programs
in AVS to apply the same phase-lag self-similarity tests  to  the
new   data.    In  addition,  their  staff  mathematician  easily
programmed new AVS modules to look for correlations  between  the
hand  motion  data  and  EEG  data  taken  simultaneously, and to
compute the Lyapunov exponents of the data streams (a measure  of
the  degree  of  mathematical "chaos" in a system).  Another more
prosaic application of this technology was simply to  record  the
hand  motions for later playback, to aid clinicians in evaluating
the efficacy of treatments and rehabilitation efforts spread over
many months [27, 28].

In addition to the above work, some of which was  part  of  their
original  research  plans,  this  team  also  was able to do some
exploratory programming, looking for new ways to visualize  their
data.   Because  of  the  ease  of  modifying AVS applications, a
number of "visual experiments" were tried, one of which  resulted
in  a useful new method for displaying and interpreting EEG data.
An  AVS  module  took  the  channels  of  data  coming  from  the
electrodes  on  the  scalp  and interpolated them -- based on the
conventional electrode positions -- into a  two-dimensional  grid
of  values.   The  time  steps  over which the data was collected
formed a third dimension, yielding a 3D volume of space-time data
from   each   EEG   recording   session.    Several  standard  3D
visualization techniques were  applied  to  these  data  volumes,
including  isosurfaces  (surfaces  of  constant value), scattered
spherical "bubbles" of varying size, and 3D gradient vectors, all
brightly color-coded.   The resulting displays made features such
as alpha-waves much more identifiable and detailed, and  provided
new  extensive  visual  cues  as to how the waves spread over the
brain  surface,  in  comparison  to  the  traditional  method  of
plotting  parallel 2D strip charts in black and white, which made
little use of the spatial information implicit in the  data.   (A
complete treatment of this innovation has not yet been published,
though  the  researchers  have  mentioned  it   in   passing   at
conferences where other research was presented. [29])

Another fortuitous avenue of research opened  up  when  the  team
obtained  digital  pressure sensor sheets, another novel computer
input  device.   With  funding  from  a  supplier  of  wheelchair
cushions,  they  embarked  on  a  study  which  yielded the first
quantitative physiometric  analysis  of  the  physical  pressures
which  cause  "chairsores"  in  wheelchair-bound  patients.  Time
sequences  of  data   collected   with   the   pressure   sensors
(affectionately  nick- named "butt-print" data) was imported into
AVS for visualization as space-time volumes in the  same  way  as
the  EEG  data, except that no interpolation was needed since the
data was already in a grid as it came  from  the  sensor  arrays.
Some  of  the  previous  work done to explore the EEG volumes was
applied to the new pressure sensor  volumes  with  good  results.
The   isosurface  technique  proved  especially  useful,  because
intervals of unchanging pressure  (which  can  cause  chairsores)
appeared  dramatically  as  straight-sided shafts of color in the
space-time volumes [30, 31].

Over time, an increasing fraction of the work done at Loma  Linda
has   been   opportunistic,   as  new  technologies  have  become
available.  The flexibility of AVS's visual programming interface
has  allowed  the  researchers  to  adapt  to these opportunities
quickly, without imposing a crushing software development  burden
as  they  race  to  keep  up  with  advances  in  the hardware of
physiometric data gathering.

REFERENCES

 1.  Warren S. McCulloch and Walter H. Pitts (Massachusetts
     Institute of Technology)  A Logical Calculus of the Ideas
     Immanent in Nervous Activity  Chapter 2 of Embodiments
     of Mind    MIT Press  1965
 2.  Michael A. Arbib (University of New South Wales)  Brains,
     Machines, and Mathematics   McGraw Hill Book Company  1964
 3.  Alan M. Turning  On Computable Numbers with an Application
     to the Entscheidungsproblem   Proceedings of the London
     Mathematical Society, series 2  42  1936  pp. 230-265
     (corrections in 43  1936-37  pp. 544-546)
 4.  The Editors of Scientific American  Information   Scientific
     Amer. Press  1966
 5.  Maurice V. Wilkes  Memoirs of a Computer Pioneer   MIT Press
     1985
 6.  Martin Cambell-Kelly  The Airy Tape: An Early Chapter in the
     History of Debugging    IEEE Annals of the History of
     Computing  14(4)  1992
 7.  David Sheff Game Over: How Nintendo Zapped an American
     Industry, Captured Your Dollars, and Enslaved Your Children
     Random House, New York  1993
 8.  Kurt Godel Uber formal unentscheidbare Satze der Principia
     Mathematica und verwandter Systeme   Monats. Math. Phys.  38
     1931  pp. 173-198
 9.  Alexander Hellemans and Bryan Bunch  The Timetables of
     Science: A Chronology of the Most Important People and
     Events In the History of Science  Simon and Schuster  1988
10.  Deborah Sojka and Philip H. Dorn Magical Moments In Software
     Datamation  ____  1981  pp. 7-16
11.  Ivan Sutherland (Harvard University)  Facilitating the Man-
     Machine Interface  Chapter III Section 1 of PURPOSIVE SYSTEMS
     Proceedings of the First Annual Symposium of the American
     Society for Cybernetics   von Foerster, White, Peterson,
     Russell, eds. (University of Illinois)  Spartan Books 1968
12.  P. Haeberli  ConMan: A Visual Programming Language for
     Interactive Computer Graphics   Computer Graphics  22(4)
     August 1988
13.  Craig Upson, Thomas Faulhaber, Jr., David Kamins, David
     Laidlaw, David Schlegel, Jeffrey Vroom, Robert Gurwitz,
     Andries van Dam (Stellar Computer Inc.)  The Application
     Visualization System: A Computational Environment for
     Scientific Visualization   IEEE Computer Graphics and 
     Applications  July 1989
14.  Larry Gelberg, David Kamins, Jeff Vroom (Stellar Computer
     Inc.)  VEX: A VOLUME EXPLORATORIUM  An Integrated Toolkit
     For Interactive Volume Visualization  Chapel Hill Volume
     Visualization Workshop 1989
15.  Marc L. Kessler, Leo R. Catallo, Dan L. McShan (University
     of Michigan, Ann Arbor)  Design and Simulation of Conformal
     Radiation Therapy Using AVS   International AVS User Group
     Conference (proceedings) 2(1) May 1993
16.  R. M. Burnett, P. L. Stewart, S. D. Fuller (Wistar
     Institute) and C. J. Mathias  (Stellar Computer Inc.)
     Image Reconstruction of Adenovirus: A relatively new
     technique may be better suited than traditional methods to
     study the structure of this complex virus  Pixel: The
     Magazine of Scientific Visualization  2(2)  July/August 1991
17.  Craig J. Mathias (Stellar Computer Inc.) [describing work of
     Phoebe L. Stewart and Roger M. Burnett of the Wistar
     Institute, Philadelphia, PA, and Stephen D. Fuller of the
     European Molecular Biological Laboratory, Heidelberg,
     Germany]  Visualization Techniques Augment Research into 
     Structure of Adenovirus   Scientific Computing and Automation
     7(6)  April 1991  pp. 51-56
18.  M. Dalton, S. Hart, A. Wolfe, M.D. (VOXEL Corp.) 
     Holographic Display of Medical Image Data   International
     AVS User Group Conference (proceedings) 2(1) May 1993
19.  Arthur J. Olson and David S. Goodsell (Scripps Research
     Institute) Visualizing Biological Molecules: Computer-
     generated images are aiding research in molecular structure
     and helping to elucidate the complex chemistry of life
     Scientific American  267(5)  November 1992  pp. 76-81
20.  Bruce S. Duncan, Michael Pique, Arthur J. Olson (Scripps
     Research Institute)  AVS For Molecular Modeling 
     International AVS User Group Conference (proceedings) 2(1)
     May 1993
21.  Gloria E. O. Borgstahl, Hans E. Parge, Michael J. Hickey,
     Robert A. Hallewell, John A. Tainer (Scripps Research
     Institute) Radical interceptor: this antioxidant may be one
     of the body's defenses against aging  (cover illustration)
     Scientific American  267(6)  December 1992  front cover and 
     p. 8 (for article on pp. 130-141)
22.  Olaf H. Weiner, R. Albrecht, B. Muhlbauer, H. Schwarz, M.
     Schleicher, A. A. Noegel (Max-Planck-Institut fur
     Biochemistry)  The Actin-binding protein comitin is a
     ubiquitous Golgi component: Analysys of its distribution by 
     confocal microscopy  Journal of Cell Science 1993
23.  K. K. Chan, A. S. Hayrapetian, C. C. Lau, R. B. Lufkin
     (University of California, Los Angeles, Department of
     Radiology)  Neural Network based Segmentation System
     Proceedings of the Society of Photo-optical Instrumentation
     Engineers (SPIE)  1898(1)  1993  pp. 604-608
24.  Alan Garfinkel (University of California, Los Angeles),
     Mark L. Spano, William D. Ditto, James N. Weiss
     Controlling Cardiac Chaos   Science  257(5074)
     28 August 1992  pp. 1230-1235 
25.  Ivars Peterson and Carol Ezzell [survey of the work of a
     number of groups]  Crazy Rhythms: Confronting the complexity
     of chaos in biological systems   Science Weekly  142(10)
     pp. 156-159  5 September 1992
26.  Xu Nan and Xu Jinghua (Academia Sinica, Shanghai, China)
     The Fractal Dimension of EEG As a Physical Measure of
     Conscious Human Brain Activities  Bulletin of Mathematical
     Biology  50(3)  1988  pp. 559-565
27.  David Warner  A Helping Hand: Computer Graphics Tools Lend
     a Hand to Victims of Parkinson's Disease  Computer Graphics
     World  October 1991
28.  David Warner [describing the research of Dr. Doug Will,
     Dave Warner, Stephen Price, Jeff Sale, Jodie Reed and Jan
     Henderson at Loma Linda University Medical Center]  Medical
     Rehabilitation, Cyber-Style: Loma Linda's Advanced
     Technology Center puts virtual reality spin-offs to work in 
     the clinic  Virtual Reality Special Report  July 1992
29.  E. Jeffrey Sale, A. Douglas Will, M.D., Jeffrey M. Tosk,
     Stephen H. Price (Loma Linda University Medical Center)
     Modifications of a Model of Chaotic Dopamine Neurodynamics
     SIAM Conference on Applications of Dynamical Systems
     October 15-19, 1992  (no proceedings published!)
30.  S. H. Price, D. J. Warner, E. J. Sale, A. D. Will, M.D.
     (Loma Linda University Medical Center)  Visualizing
     Physiological Data -or- How To Teach Doctors New Tricks
     With AVS   International AVS User Group Conference
     (proceedings) 2(1) May 1993
31.  Janet L. Henderson, MD, OTR, Stephen H. Price, BS, Murray E. 
     Brandstater, MBBS, PhD, Benjamin R. Mandac, MD (Loma Linda
     University Department of Physical Medicine and
     Rehabilitation)   Efficacy of Three Measures to Relieve
     Pressure in Seated Persons With Spinal Cord Injury  
     Archives of Physical Medicine and Rehabilitation (preprint;
     accepted for publication in early 1994)

Return to to ABS's home page.