Killam Award: Mining data to save lives

In some cases, finding the needle of knowledge in a haystack of data can mean the difference between life and death. No one knows that better than Killam recipient and professor in the Department of Computing Science Osmar Zaïane, who applies his considerable knowledge of data mining to problems in health care.

Data mining, or knowledge discovery, involves sifting through vast stores of information—too large for human beings to manage without computer aid—to find connections and patterns that are of practical use. A simple example is the feature on Amazon.com that provides customers with a personalized list of titles they might be interested in buying, based on previous purchases.

But while helpful in marketing, data mining has far more critical applications. One of Zaïane’s current projects with the Cross Cancer Institute, for example, is aimed at improving the interpretation of mammograms.

“There are thousands and thousands of mammograms that doctors have to go through each year,” he says. “They are difficult to analyze, and humans make mistakes.” In fact the error rate for mammograms is estimated at about 15 per cent.

To reduce that, it’s recommended that two doctors read every mammogram, says Zaïane, but with a shortage of radiologists, only about 10 per cent of mammograms in Alberta are randomly selected for a second opinion. That’s where data mining can make a big difference.

“Why select randomly?” asks Zaïane. “With machine-learning techniques, we build a model of what type of cancer is in a mammogram. So, based on examples radiologists have already annotated, we can label a mammogram as indicating cancer with some certainty. There is confidence attached to the prediction we make.”

Zaïane says it’s not about replacing the human expert in making a diagnosis, but rather narrowing down the pool selected for second reading with intelligent criteria. “When the machine can’t make a prediction with any degree of certainly, then maybe those are the cases that should be sent for second reading [by a doctor]. Rather than selecting randomly, we select, in a smart way, the cases that are difficult.”

The prototype Zaïane and his team have designed is 82 per cent accurate so far, only three per cent short of the average human rate of success. And since the computer learns as it acquires more information, including feedback from doctors, it only gets better.

But the uses for data mining in health care don’t stop there, says Zaïane. He’s also working on a prostate cancer project, again with the Cross Cancer Institute, comparing genes to determine how patients will react to treatment. This kind of genetic analysis can also be used to determine whether a patient will reject a transplanted organ.

Having been with the U of A for only 10 years and already regarded as an innovator in his field, Zaïane is described as a “a rising star” with a publication record “eclipsing many of our full professors,” says Jonathan Schaeffer, former chair of computing science. Schaeffer adds that Zaïane has also built an impressive multidisciplinary research program with a team of dedicated graduate students and recently took over leadership of the multi-million-dollar Alberta Ingenuity Centre for Machine Learning, which is focused on health and medical informatics.

“Osmar is a dynamic individual, with seemingly boundless energy,” says Schaeffer. “He is an excellent researcher, teacher, a superb mentor of students, a committed teacher and a valued colleague.” 

2010–2011 Killam Annual Professorships

Applications are invited for the 2010–2011 Killam Annual Professorships. All regular, continuing, full-time academic faculty members who are not on leave during 2010–2011 are eligible to apply. Deans, department chairs and other senior university administrators with personnel responsibilities shall not normally be eligible for Killam Annual Professorships. Associate deans and associate department chairs are eligible, providing they do not have personnel responsibilities. Up to eight Killam Annual Professors will be selected by a subcommittee of the Killam Trusts Committee; no more than two professorships shall be awarded to staff members in any one faculty in any given year. Each Killam Annual Professor shall be presented with a $3,500 prize and a commemorative plaque. The duties of Killam Annual Professors shall not be changed from those that they regularly perform as academic staff members.

The primary criterion for selection shall be a record of outstanding scholarship and teaching over three or more years as evidenced by any or all of research publications, creative activities, presented papers, supervision of graduate students and courses taught. The secondary criterion shall be a record of substantial contributions to the community outside the university, above and beyond what is usually expected of a professor.

Awards are tenable for 12 months commencing July 1. The completed application must be received at the Office of the Vice-President (Research), 203 TELUS Centre, by 4:30 p.m., Friday, Feb. 19. The awardees shall be announced by early May, and they will be formally recognized at the Killam Luncheon in the fall of 2010.

Applications and further details are available on the home page of the Vice-President (Research) at: www.uofaweb.ualberta.ca/vpresearch

Please contact Annette Kujda, administrative officer, Office of the Vice-President (Research) at 780-492-8342 or email: annette.kujda@ualberta.ca if you have any questions.

Next: Vaccine blitz a success