Oak Ridge National Laboratory



Media Contact: Ron Walli (wallira@ornl.gov)
Communications and External Relations


ORNL helping surgeons, patients

OAK RIDGE, Tenn., Feb. 2, 2001 — Math and medicine are coming together to help people who have suffered an abdominal aortic aneurysm, which with 15,000 is the 13th-leading cause of death in the United States.

At the heart of the effort are genetic algorithms written by Oak Ridge National Laboratory researchers that allow physicians to more efficiently assess and organize the often vast amounts of information contained in patient reports. Ultimately, with this tool - a sophisticated way to quickly extract key phrases - doctors will be able to characterize features and findings in reports and provide better patient care.

"If a surgeon had a way to more accurately predict whether a patient is likely to suffer a leak or rupture, he could be in a better position to help that patient," said Robert Patton of ORNL's Computational Sciences and Engineering Division. "We believe our method to quickly extract and organize information from reports will be a huge asset to surgeons and their patients."

An abdominal aortic aneurysm occurs when the artery that supplies blood to the abdomen expands under pressure or balloons outward. Most occur in males 60 or older with risk factors including tobacco use, high blood pressure, atherosclerosis and race as this is most common in whites, according to the Mayo Clinic (http://www.mayoclinic.com/).

This work builds on previous studies involving genetic algorithms developed for mammography. That system allows doctors to quickly identify trends specific to an individual patient and match images and text to a database of known cancerous and pre-cancerous conditions.

In much the same way, Patton and colleagues see this benefitting people who have had to undergo surgical repairs for an abdominal aortic aneurysm - usually with the insertion of a graph or stent. After the surgery, physicians typically monitor the patient for several years to ensure that there are no further ruptures or leaks.

For the study, researchers examined records of 20 patients and a total of 111 reports consisting of 87 radiology reports and 24 operative notes. The operative notes are comprised of approximately four radiology reports and one operative note per patient. Unlike the mammography reports that were the focus of a previous study, reports for aortic aneurysm patients had far more variability in the language and tended to be longer. They lacked the frequent labels "normal" or "suspicious" found in mammogram reports.

Researchers noted, however, that patient reports with abnormalities tended to be longer and contained wider variation in the language than those of normal patient reports. These reports also contained more "negation" phrases such as "no evidence of endograft leak."

"Because of the length and variability of the reports, the task of retrieving just reports that represent abnormalities is daunting," Patton said.

Consequently, the conventional approach using vector space model based on individual terms does not adequately capture the language used in these medical documents. To address this challenge, the team employed what's known as "skip grams," which are word pairs in their respective sentence order that allow for different gaps between the two words. For example, the researchers may select "no" and "leak" as one of the sought parameters, but the level of sophistication can be increased dramatically.

By employing this strategy, the researchers were able to extend and apply the Maximum Variation Sampling Genetic Algorithm to the radiology reports.

While this work has been presented at workshops and published in MedGEC, a follow-on paper about type II endoleaks has been submitted to the annual conference of the Society of Vascular Surgery. Researchers involved in the abdominal aortic aneurysm work include Barbara Beckerman and Vincent Paquit of ORNL's Computational Sciences and Engineering and Measurement Science and Systems Engineering divisions, respectively.

This research was funded by ORNL's Laboratory Directed Research and Development program and made possible by the University of Tennessee Graduate School of Medicine and University of Tennessee Medical Center.