|
Cover Story Sidebar |
|
|
Mining Breast Cancer Imaging Data
Software will provide key indicators by JANE M. SANDERS | |
IN THE SECONDS it takes to do a Google search, a software tool under development at Georgia Tech could provide key indicators drawn from medical imaging data to help doctors making a diagnosis and prognosis of breast cancer.
|
The image-enabled data mining system, which could be expanded to find other diseases, would be a support tool for breast-imaging radiologists, as well as medical researchers. The work is being supported, in part, by the Georgia Cancer Coalition (GCC).
“The software could help a radiologist better diagnose breast cancer,” says Christopher Barnes, an associate professor of electrical and computer engineering in Georgia Tech’s Regional Engineering Program in Savannah, Ga. “If the doctor finds a suspicious area on a mammogram, this software tool might help the physician decide what to do next and how urgently action is required.”
Barnes’ software would find relevant case histories over the entire archived database – even one including millions of patient records.
“This tool would look at the radiologist’s region of concern in the current patient’s mammogram and, using the mammogram’s pixels as a query, determine all relevant, on-line case histories – that is, others who have had suspicious mammograms with similar micro-calcification clusters,” he explains. “The purpose is to find relevant case histories – with their recorded indicators, diagnosis, prognosis and outcomes – to support a diagnosis and post-diagnosis decisions for the current patient.”
Barnes has gotten input from Emory University breast imaging expert Dr. Carl D’Orsi on the key information that physicians need from a database query.
The front end of the system (i.e., the doctor’s interface) would run on a desktop computer, while the database would reside on another, more powerful computing system. The doctor would submit an image of the suspicious portion of a patient’s mammogram as a query to the clinic’s archive.
“The tool would help answer questions like: ‘What is the likelihood of a malignancy based on all the relevant case histories? Is there enough concern to order a biopsy?’ We’re trying to reduce the rate of unnecessary biopsies,” Barnes explains.
Shown on this mammogram is a small cancerous lesion indicated by an arrow, as well as calcific deposits in the veins.
“Right now, as many as four in five biopsies prove a patient to be cancer free,” he adds. “Although a cancer-free diagnosis is wonderful news, the doctor’s order for a biopsy incurs medical risks, emotional trauma and unnecessary expense. They are ordered based on a doctor’s review of the patient’s mammogram. We’d like to get the biopsy rate down to two or three in five by helping the radiologist use comparative analysis tools. The software won’t be successful unless it can improve the unnecessary biopsy rate without increasing the missed detection rate. That is the goal for the level of care in a clinic.”
A doctor might also use this software to determine what treatment methods are most suitable based on relevant case histories. “The query results might indicate which ordered procedures have resulted in the past in the best outcomes,” Barnes explains.
“An experienced doctor may not learn as much from the database search as a new doctor,” Barnes adds. “But this tool would equalize the care provided across a clinic.”
Barnes’ software also has a clinical research application because of its potential ability to search based on multiple criteria, including imagery, patient information, diagnosis, treatment procedures, outcome history and even genomic-related clinical data, including DNA microarray results, which might indicate specific sub-types of cancer.
“Researchers want to know what genomic profiles are most at risk and what mammographic indicators might relate to these genetic profiles,” Barnes says. “They might discover possible correlations between the two. For example, a doctor might find both an environmental and DNA-based risk of breast cancer.”
Barnes began development of his software in the mid-1990s at the Georgia Tech Research Institute. The GCC grant awarded in 2003 will help him move the software toward clinical trials. Barnes believes one key advantage of his software will be its acceptance by doctors and patients. Unlike some commercial programs available, Barnes’ software does not attempt to detect cancer on its own, but instead provides comparative information to help the physician make the diagnosis and prescribe treatment.
“I believe this software is going to make a real difference,” Barnes adds.
courtesy National Cancer Institute
For more information, contact Chris Barnes at 912-966-7927 or
chris.barnes@gtrep.gatech.edu.
Contents
Research Horizons
GT Research News
GTRI
Georgia Tech
Send questions and comments regarding these pages to
webadmin@edi.gatech.edu
Last updated: July 7, 2004