The quantifier: Building software that interprets medical images

January 12, 2012 by Larry Hardesty

Polina Golland’s parents tell the story that, sometime in the early 1980s, when Polina was in junior high, she announced that she wanted to go to MIT. That’s an unusual plan for any 13- or 14-year-old to hatch independently, but particularly one living in the Soviet Union during the Cold War.

Golland, now a newly tenured associate professor in the Department of Electrical Engineering and Computer Science and a principal investigator at the Computer Science and Artificial Intelligence Laboratory (CSAIL), grew up in Frunze (since renamed Bishkek), the capital of Kyrgyzstan, a city of about a million people. “It was actually European in nature,” Golland says. “It was mainly Russians.” Golland had what she calls a “happy childhood,” dominated by physics and math. Both of her parents were physics researchers — her father at a Kyrgyz university and her mother in the National Academy of Sciences — and she was enrolled at the Frunze science school.

Golland doesn’t recall the incident, but according to her parents, her ambition to attend MIT awoke when she pulled a volume of the celebrated Lectures on Physics by Richard Feynman ’39 off her father’s bookshelf. In a stroke of luck, the volume was one that, as the jacket copy declared, Feynman had written while on sabbatical at MIT. If Golland had happened to grab a different volume — who knows, she might have ended up at Caltech.

As an undergrad, however, she initially enrolled at a university in Moscow, where she got her first taste of the Soviet police state. “Some of my friends in college were called to the KGB and asked to report on what was going on in their discussions,” Golland says. “I didn’t particularly like that.” When, in 1990 — after the fall of the Berlin Wall but before the dissolution of the U.S.S.R. — her parents floated the idea of emigrating to Israel, Golland and her sister were all for it, even though it meant renouncing their Russian citizenship.

After earning both bachelor’s and master’s degrees in computer science at Israel’s Technion Institute of Technology, Golland finally came to MIT for her doctorate — not in fulfillment of a childhood prophesy but simply, she says, “because that’s the best school to do a PhD at.”

Now a citizen of both the United States and Israel, Golland has been back to Russia only once since 1990. “I feel a lot of connection to Russian culture,” she says. “I speak Russian to my 2-year-old, but I don’t feel a connection to the country. I sort of view myself as Israeli-American.”

Golland describes her educational background as “ with an unhealthy dose of math”; during her final semester at the Technion, she took a course whose mathematical rigor appealed to her so much that she decided to do her master’s research with the instructor. The course happened to be on image processing, and for her thesis, she developed techniques for extracting information about the motion of objects in digital video — a problem squarely in the mainstream of computer-vision research.

For her PhD work, she had anticipated doing something similar. But when she arrived at MIT, she found that some of her new colleagues were working on algorithms for processing medical-imaging data, and she was intrigued. “Most of my students haven’t seen a biomedical image in college,” she says. “I get them interested in this area, oftentimes after they are admitted. And the same thing happened to me.” For her dissertation, Golland worked on algorithms for identifying variations in the shape of the hippocampus that were correlated with neurological disease.

As a professor, Golland remains interested in anatomical variation, but in a broader range of contexts. How can an algorithm learn to automatically identify organs in medical images, given that the images used to train the algorithm will depict organs with subtly different shapes? Can a mathematical model of organ shape be made flexible enough to identify organs that have been grossly distorted — say, by the pressure of a tumor? How do organs change shape as a disease progresses?

She’s also branched out to another major area of research: modeling brain activity. Functional magnetic resonance imaging (fMRI), which can measure oxygen levels in animal tissue, has been used to identify regions of the brain that receive a rush of blood when test subjects are asked to perform certain cognitive tasks. But many complex tasks, rather than sparking a lot of activity in one region, spark subtle activity in a lot of different regions, which can be very difficult to distinguish from the background hum of the resting brain.

In such cases, simply turning a computer loose on raw fMRI data and asking it to look for patterns would be fruitless. So Golland’s group is working closely with a number of neuroscientists, who provide working hypotheses about brain activity that serve as the basis for mathematical models. These models tend to bring the hypotheses into sharper focus, which leads to their revision, which leads to new models, and so on, in a continual process of iteration. “I love that process,” Golland says. “They refine their notions of what they’re trying to formulate, and we understand much more about what’s known about the biology.

And that points to one of the reasons that analyzing appealed to Golland so much in the first place. “It’s lots of fun, because I get to stay a student all the time,” she says. “By the nature of it, we always have to keep learning about it.”

Explore further: The kids are alright

Related Stories

The kids are alright

May 26, 2011

Children should be seen and not heard... who says? A Philosophy academic at The University of Nottingham is challenging the adage by teaching primary school children to argue properly.

Is there a hidden bias against creativity?

November 18, 2011

CEOs, teachers, and leaders claim they want creative ideas to solve problems. But creative ideas are rejected all the time. A new study, which will be published in an upcoming issue of Psychological Science, a journal of ...

How your brain reacts to mistakes depends on your mindset

September 30, 2011

(Medical Xpress) -- “Whether you think you can or think you can't -- you're right,” said Henry Ford. A new study, to be published in an upcoming issue of Psychological Science, a journal of the Association for Psychological ...

Sexism and gender inequality

October 28, 2011

(Medical Xpress) -- Individual beliefs don’t stay confined to the person who has them; they can affect how a society functions. A new study published in Psychological Science, a journal of the Association for Psychological ...

A call for an evolved understanding of emotion

January 4, 2012

(Medical Xpress) -- Many scientists believe that all people experience and express the same biologically “basic” emotions — an idea they have attributed to evolutionist Charles Darwin and one that has shaped ...

Recommended for you

Dutch open 'world's first 3D-printed bridge'

October 17, 2017

Dutch officials toasted on Tuesday the opening of what is being called the world's first 3D-printed concrete bridge, which is primarily meant to be used by cyclists.


Adjust slider to filter visible comments by rank

Display comments: newest first

not rated yet Jan 12, 2012
How can an algorithm learn to automatically identify organs in medical images, given that the images used to train the algorithm will depict organs with subtly different shapes?

As someone who has worked on this himself I can say. There's a wide gap between the mathematically feasible and the medically useful. Atlas based training works great for segmentation of healthy structures. However, in medical contexts you are never interested in healthy structures. Especially lacerations, fractures and other, major pathologies are a real problem.
They manifest in many ways - and almost always have a very unique look to them in medical images. So generalized approaches based on shape factors alone are pretty much doomed to failure except for the most trivial cases.

Texture based approaches have, in my exprience, been much more successful at reliably identifying pathological for general cases than shape based ones.
not rated yet Jan 13, 2012
This article makes no sense. Apart from assuring the reader that M.I.T. is the best place for science in the Free World, and prof. Golland considers herself Israeli-US with absolutely no real connection to Russia (except for appreciation for its very, very exotic culture) it doesn't contain almost anything on prof. Golland's research.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.