Pre-med, to 1970:
I became interested in computers while still in high school, reading
some of the manuals for an IBM 360 during my job as a file clerk in an
insurance office (the file clerk job didn't exactly turn me on to insurance,
but it did turn me on to computers). In college I learned FORTRAN as my
first computer language, spending more time programming an IBM 1130
for physics assignments than I spent understanding the physics.
Medical school, 1970-1974:
Medical school washed away a lot of my math and computer skills,
but I rediscovered computers when I took an elective in 1974 at NIH, called
"Computers in Clinical Medicine". This course was taught by Bill Mohler
at the Division of Computer Research and Technology (DCRT), and was a precursor
to a similar elective now offered by NLM. I credit Bill Mohler and
this course with really opening my eyes to the possibilities of computers
in medicine, and letting me realize that I could combine my interest in
computers with my background in medicine.
Postdoctoral fellowship, 1974-1977:
Following medical school I decided I was more interested in computers
than medicine, so I worked as a postdoctoral fellow in the University of
Washington Department of Bioengineering, with Don Baker, Bill Moritz, Ron
Daigle and Dave Phillips, among others. The project was the development
of methods for computing organ volume using 3-D ultrasound. Since 2-D ultrasound
scanners had only just become available I spent quite a bit of time developing
methods for computing the 3-D location of a 2-D scan plane in space, using
a spark gap position locating system developed by Moritz and his students.
Ph.D., 1977-1984:
During the UW postdoc I realized that I really wanted to learn
all I could about computers, so I enrolled in an interdisciplinary Ph.D.
program at Stanford, where I developed a course of study that combined
computer science, medicine and engineering. I called this program, "Medical
computer engineering". Nowadays it would be called medical informatics.
My advisors were Gio Wiederhold and Tom Binford from CS, Al Macovski from
EE, Rich Popp from Cardiology and Desmond McCallum from Ob/Gyn. I took
all the courses taken by CS Ph.D. students, and did my research in Ob/Gyn
with Des McCallum. There I adapted my 3-D ultrasound work to develop methods
for measuring 3-D volume as a predictor of fetal weight. I also applied
some of my knowledge of image understanding to develop a shape based approach
to organ segmentation from 3-D ultrasound images. This work because my
Ph.D. thesis, "Ultrasonic Three-dimensional Organ Modeling", and was the
basis for most of my early papers and book chapters. I believe it was some
of the earliest work in 3-D ultrasound, a field that has currently become
very popular.
Research Associate, 1984-1989:
Following my Ph.D. I worked for Bruce Buchanan as a research associate
in the Stanford Knowledge Systems Lab, which is part of the Department
of Computer Science. While there I was a member of the PROTEAN project,
whose other members included Bruce Buchanan, Oleg Jardetzky, Russ Altman,
Bruce Duncan, Barbara Hayes-Roth, Mike Hewett, Olivier Lichtarge, and Craig
Cornelius, among others. The problem we were trying to solve was the determination
of 3-D protein structure from nuclear magnetic resonance spectroscopy (NMR),
using artificial intelligence techniques to limit the very large search
space. My main contribution was the formulation of the problem as a geometric
constraint satisfaction problem, based on my earlier work in ultrasonic
organ modeling. I also designed and largely implemented GS1, (geometry
system 1), a C-based Internet server for solving geometric constraint satisfaction
problems. BB1, a blackboard based problem solving system developed under
the direction of Barbara Hayes-Roth, was then used to control GS1
over the net.
Research Faculty, 1989-present:
I was recruited to the UW Department of Biological Structure by Cornelius
Rosse and John Prothero, who had been involved for several years in 3-D
reconstruction of organs from serial sections. Based on a newly funded
grant from NLM, they were interested in developing a knowledge base of
anatomy that could serve to organize the 3-D models and to provide a basis
for organizing other information. My initial contribution was to describe
anatomical (and more broadly, structural) knowledge as being composed of
two parts, a spatial component and a symbolic component. Based on my Stanford
experience I then designed a distributed, Internet-accessible architecture
for representing and accessing these two kinds of knowledge. This framework
has proven to be very resilient, and continues to be our main architecture
ten years after it was designed.
During this time I also generalized the shape-based imaging work and the protein structure work to come up with the notion of hierarchical geometric constraint networks as a representation for spatial structural knowledge, at levels ranging from gross anatomy to molecules. The basic notion is that networks of local geometric constraints, when interacting together in a constraint satisfaction process, give rise to a global representation of shape, for both individual structures and for collections of structures.
In 1990 I coined the term, "Structural Informatics", partly to give a name to the kind of activities I had been doing since 1974, but also in recognition of the commonality of representations and methods for structures at all levels from gross anatomy to molecules. This term has since been picked up by NLM as a subfield of medical informatics. In 1997 the name of our group was changed to the Structural Informatics Group. I became director of the group in 1997, in close collaboration with Cornelius Rosse, the original founder of the Digital Anatomist project.
Together with Cornelius Rosse, I now direct a group of students and programmers in the development of computer-based representations and applications in anatomy. Whereas Cornelius concentrates more on anatomical knowledge representation, I concentrate more on the design and implementation of computer-based methods for organizing, accessing, visualizing and utilizing the knowledge that he represents. A good team, I think.
In the long run I would like to work with Cornelius and other domain experts to develop representations and software tools that can work together over the Internet, and that can lead to a structural information framework for organizing and accessing biomedical information. Because of the fundamental nature of structure, such a framework has the potential to organize a large portion of biomedical information, the need for which is becoming apparent as the amount of information continues to exponentially increase.