Steven Simske's "real" job is finding ways to improve the extraction of useful information from electronic sources -- documents, repositories, collections containing materials such as text, images and/or speech.
But that expertise, plus a degree in biomedical engineering, has lead him in some unexpected directions. Recently, he's been working medical researchers to allow them to extract data from images
Simske joined HP Labs in 1999 after working for several years on document understanding for HP's scanner division.
We interviewed him recently and asked him to talk more about his work.
What is the primary focus of your research?
I am looking at how to improve content extraction and upgrade using combinations of algorithms and/or services. I am also looking into how to set up dynamic combinations of these algorithms and services to minimize machine intelligence errors and to optimize machine responsiveness to user input (as speech, text, scanned documents, images, etc.). It may seem a bit broad, but "perfecting" the extraction of information that humans can readily ascertain from various electronic sources (text, image, speech, etc.) is the common thread.
How did you get involved in medical image understanding?
I did my Masters degree work at Rensselaer, in impedance tomography (a means of imaging using currents applied to a body surface and subsequent recordings of the voltages). In addition to my work at Rensselaer, I worked on bone histomorphometry, collagen and fibrin assembly analysis, implant ingrowth and apposition and microhardness at the University of Colorado.
What exactly are you working on?
We're trying to allow new image-understanding tasks to be performed in a dynamic manner. In medical imaging, if you have great big systems that do all of the image analysis, you end up with something that's just a mess. However, if you write simple little bits of software that do specific tasks on an image, such as finding regions of the same color or doing edge detection (segmenting images into their components, or regions of interest, such as separating faces from the rest of the body) you can piece all of these together dynamically. This allows you to create overall applications that you couldn't have planned for in advance.
I'll try to explain this with an example: If you wanted to find all of the spots on a set of slides that might be indicative of breast cancer, you may not have written a specific software algorithm that knows what breast cancer is. But what you can do is piece together algorithms for edge detection, region-finding, thresholds (algorithms that affect what is declared "foreground" and "background") and so on, and then check the color of those regions to see if they look like a cancerous area. The user should be able to use natural language commands, which the machine can convert into executable tasks.
Edge detection region-finding and threshold algorithms are generic algorithms used for image understanding. We are concerned with dynamically linking these algorithms to solve tasks not envisioned a priori (and not envisioned by the algorithms themselves).
Is there interest from the medical community in this research?
We're working with the Webb-Waring Institute Institute for Cancer, Aging and Antioxidant Research in Denver -- their director wanted to know if it was possible to write software that can perform the task of detecting glaucoma. To do this, you can either write a specific piece of software or you can write a series of algorithms, which can be used for detecting other diseases too. Webb-Waring have asked us to look at a couple of things for them: investigating the formation of macular degeneration and the detection of breast cancer.
What is the status of that work?
Nascent. We are currently focusing on shape and facial recognition. We are also looking at extracting quantitative information about various bone implants -- bone ingrowth, bone apposition (growth against the implant), etc.--including titanium oxide and tricalcium phosphate porous implants; and for Mitek anchors, which anchor tendon and ligament protheses into the bone. The latter is with the Colorado School of Mines. Webb-Waring is also interested in the "quantitative histomorphometry" to automate their histology tasks (studying cells at the microscopic level).
You're also involved in research into lifetime hip implants
Yes, I'm working with the Colorado School of Mines on this. Typically hip implants need to be replaced after 12 to 15 years because they're solid, and the bone that's up against the implant disappears. It therefore makes sense for a lifetime implant to be porous so that the bone can grow through it. We can add value to this research by automating tasks and providing quantitative data about cell counts and types that is otherwise just too time-consuming to obtain.
What about your work on preventing osteoporosis?
We're looking at mouse models for osteoporosis, including space flight and old age. We study the effect of cytokines such as M-CSF, IGF-1, Il-1ra, and OPG on these models and on mouse development. We're also looking at suitable bone implants.
You're also flying mice and rats on the Space Shuttle?
So how does it help to send mice into space?
Unlike humans, the skeletal mass of mice and rats never stops growing. Even over a week-long space flight, a huge change in growth can occur, so this gives us a great deal of quantitative data to work with. Enough of the effects on mice can be related to humans so that the results from these experiments can make a significant contribution to our osteoporosis research.
What research are you and your team conducting in this area?
In December 2001 we flew mice in space for about 12 days, and treated them with a novel bone cytokine, osteoprotegerin. It prevents loss of bone in space mice. This work is to be published later with this year with Amgen.
What's your educational background?
My background is in biomedical engineering, which I studied at Marquette University (BS) and Rensselaer Polytechnic in New York (MS). Following my Masters, I moved on to a PhD in Electrical Engineering at the University of Colorado.
What do you do outside work?
I play sports, but with disastrous consequences at times! My nose was broken recently in indoor soccer, and my shoulder separated in 2000 playing hockey. These days I don't have as much time for sport, because I have family commitments, but I'm more than happy with that.