UT Researchers Try Computers to Tailor Health Treatments

Science’s crystal ball has always been a little foggy. But researchers at UT are working on getting it polished up.

Tinsley Oden, director of the Institute for Computational Engineering and Sciences, or ICES, presented an overview of his and his colleagues’ advances in the field on Tuesday evening at the AT&T Executive Education & Conference Center.

Scientists figured out fairly soon after the advent of the first computers that they would make pretty good fortune tellers. Write a program that takes in a bunch of data, shoot those numbers through a massive amount of computation, and out comes a prediction, faster than Ms. Cleo can read your palm.

Computers have actually been modeling the future as far back as the 1940s, when Manhattan Project researchers simulated nuclear detonations before the first bomb test actually took place.

The field has grown since. Wings stay on airplanes, buildings survive earthquakes, and passengers walk away from car crashes, all thanks in some part to science’s ability to predict the future, or at least predict the outcome of a chain of events given some initial information. ICES works to expand the realms where predicative science can be of use.

Take, for instance, the doctor. These days, treatment is derived from some general idea about how most people will progress through some disease, and how most people have reacted to this or that procedure or drug.

But what if treatment could be individually tailored to the patient? What if a doctor, with the aid of a computer program and some advanced imaging and diagnostic equipment, could make a prediction about his exact illness, and furthermore, instead of just basing the expected outcome of his medical care on general experience, actually simulate his treatment and choose the best course of action?

Oden described an ICES program trying to achieve just that. The Cardiovascular Engineering Group seeks to develop models and tools for simulating heart disease intervention outcomes not just for the population in general, but for individual patients, based on their unique circumstances. Advancements in MRI and other imaging technologies provide a trove of information about a patient, but it’s only as useful as the model, or computer program, that interprets the data.

That’s what ICES works on. But figuring out how to crunch data to make predictions is only half the battle. What data should we look at in the first place? How accurate does that data need to be to give us useful predictions? A lot of questions have to be answered before a MacBook can become Zoltar.

And science hasn’t had a perfect track record on predicting the future. Anyone who has delighted at the weatherman’s announcement of a snow day, only to deflate at the sight of a rainy morning, knows this. Oden does not deny it either. “Science does not uncover the truth,” he says. “It uncovers our best ideas about the truth, at that particular moment.”

But if science doesn’t reveal objective truth, how can we rely on it to make predictions, or more importantly, decisions, involving things as important as climate change or a person’s health?

Oden and his colleagues use a framework to embrace the untruth, he said. It’s based on Bayesian statistics, a paradigm where information isn’t considered known, but rather only probably known. That is, you don’t say it’s 105 degrees outside, but only that you’re 95 percent sure that it’s 105, because your thermometer might be a little off, or maybe sitting directly in the sun.

The funny thing is that Bayesian statistics has been around for hundreds of years. Bayes himself was born in 1701.

“This was like the Da Vinci Code for us,” says Oden. “It was literally right in front of us for 249 years.” By accounting for, to borrow a phrase, known unknowns (and maybe your unknown unknowns as well), the researchers at ICES hope to not only improve the reliability of science’s predictive models, but also the scope of where they can be of use.

Some work from ICES has already been put to practical use. Researchers at UT fed satellite pictures, wind, weather, and tide data into their models to predict the movement of oil during the BP spill, up to 72 hours in advance.

Maybe next time you’re waiting for snow, you won’t be let down.

Oden’s presentation was part of a series of monthly lectures and presentations put on at the AT&T conference center by The Austin Forum. It’s open to the public, and you can learn more at

Texture based volume visualization of the CT imaging of Visible Human data set, centered around the abdominal region. Image courtesy Computational Visualization Center/ICES.


No comments

Be the first one to leave a comment.

Post a Comment