Donate to Science & Enterprise

S&E on Mastodon

S&E on LinkedIn

S&E on Flipboard

Please share Science & Enterprise

Crowdsourcing Yields Heart Disease Algorithm

Heart in rib cage illustration


30 March 2016. Two financial analysts are the winners of a data science competition to write an algorithm that quickly analyzes MRI images of a person’s heart suspected of cardiac disease. The winning algorithm, by Qi Liu and Tencia Lee, was submitted in the second Data Science Bowl, put on by consulting company Booz-Allen Hamilton in McLean, Virginia and data science community platform Kaggle that hosted the competition.

According to Centers for Disease Control and Prevention, heart disease is the leading cause of death for men and women in the U.S., with more than 600,000 people dying each year, or 1 in every 4 deaths. Coronary heart disease is the most common type of heart disease, leading to 370,000 deaths each year. In addition, heart attacks strike 735,000 Americans annually, with most of those heart attacks — 525,000 — happening to people for the first time.

The competition, begun in December 2015, sought a faster way of diagnosing heart disease from magnetic resonance images or MRI scans. MRI scans, say the contest sponsors, are considered the best method of determining heart functions, particularly measuring the percentage of blood ejected from the left ventricle, known as ejection fraction, and volumes of blood pumped with each heart beat.

While MRI scans provide good raw material for those measurements, the process of determining the presence of heart disease from MRI scans is slow and manual. A trained cardiologist needs to review the images and make the calculations, a process taking about 20 minutes, according to contest organizers.

The Data Science Bowl asked contestants to write an algorithm that analyzes MRI images of the heart and makes the measurements now requiring trained staff. National Heart, Lung, and Blood Institute at National Institutes of Health and Children’s National Medical Center in Washington, D.C. provided a set of more than 1,000 MRI images for participants to review and analyze. The competition drew 293 participants, yielding 1,392 algorithm entries.

Crowdsourcing, by definition, seeks ideas from outside the usual community of experts, and this year’s winners of the Data Science Bowl are a good example. Liu and Lee are financial analysts, known as “quants” in the trade for their skill in devising complex mathematical models for financial transactions. For this task, they turned their skills to an algorithm that employs deep learning, an artificial intelligence technique for machines to discern underlying patterns in relationships, and build those relationships into knowledge bases.

“What drew me to the challenges was its inherent complexity,” says Lee in a Booz-Allen Hamilton statement. “The data set had a lot of quirks that required us to think through unique scenarios and redirect our algorithm multiple times.”

Lee and Liu are now eligible for the first-place prize of $125,000. A team from Ghent University in the Netherlands received the second-place award that can win $50,000, and third place was awarded to freelance software developer Julian DeWit with a prize of $25,000. Researchers at NIH are first expected to test the accuracy of the algorithms. The top three entrants will also present their techniques and results next month at the GPU Technology Conference in San Jose, California.

Read more:

*     *     *

Comments are closed.