Natural Selection, Inc. Completes 12-Month Study on Computer-Aided Diagnosis of Breast Cancer: Order-of-Magnitude Improvement Attained through Evolutionary Programming

La Jolla, CA - Natural Selection, Inc. has completed a 12-month study investigating the feasibility of using computers to assist radiologists in detecting and diagnosing breast cancer from film screen mammograms. The research, sponsored by a grant from the U.S. Army Medical Research and Materiel Command†, focused on the identification of important features found in mammograms and their correlation with malignancy or benign conditions. These features were used as inputs to artificial neural networks, computer models loosely based on how neurons in a brain are connected and respond to stimuli. Networks were trained on features from more than 200 mammograms using an algorithm that simulates natural evolution. Populations of neural networks competed for survival based on their ability to assess the identified features and the presence or absence of cancer in the patient. Through hundreds of iterations of variation and selection, superior networks were evolved with a demonstrated performance on novel cases that matched the best known results in scientific literature under similar conditions. The final evolved networks, however, were as much as an order-of-magnitude less complex than previous networks. This not only reduces the computation required to generate a diagnosis, but also gives hope that the decisions made by the neural networks will be explainable.

Dr. David B. Fogel, executive vice president and chief scientist of Natural Selection, Inc., was principal investigator: "This study is an important step toward a completely automated system that will help radiologists detect malignancies earlier and reduce the number of unnecessary biopsies." Reported cancer statistics show that there are now more than 180,000 new cases and 45,000 deaths annually in the United States alone, but the field of application for computer-aided diagnosis extends beyond our own borders: "This technology affords the possibility for telemammography in remote locations where women may not have routine access to desirable levels of health care," Fogel said.

Assisting Dr. Fogel was Eugene C. Wasson III, M.D., chief of radiology at Maui Memorial Hospital, Maui, Hawaii: "We believe this will lead to more rapid, accurate, and cost-effective assessment of mammograms thus allowing for breast cancer detection as early as possible." The technology has important commercial aspects as well. Edward M. Boughton, president of Hawaii Industrial Laboratories, Inc. and former president of the Maui Economic Development Board remarked that "joining the eventual power of this screening tool with the new generation of digital X-ray systems will accelerate the entire process of mammography while simultaneously gaining confidence in the outcome." The results of the research were presented at the "Breast Cancer: An Era of Hope" conference in Washington, D.C. which was held Oct. 31 through Nov. 3, 1997. Natural Selection, Inc. continues to pursue basic and applied research in this important area of treating breast cancer through early detection and diagnosis.

Natural Selection, Inc. was founded in 1993 to address complex problems in industry, medicine, and defense. The company possesses unique expertise in computational intelligence techniques, including evolutionary computation, neural networks, and fuzzy logic. The corporation's current research efforts support the discovery of new pharmaceuticals, the automated detection and diagnosis of breast cancer from film-screen mammograms, optical character recognition, and a variety of military projects.

For more information, please contact:

Dr. David B. Fogel,
Executive Vice President & Chief Scientist,
Natural Selection, Inc.,
3333 N. Torrey Pines Ct.,
Suite 200,
La Jolla, CA, 92037
tel: (619) 455-6449
fax: (619) 455-1560
dfogel@natural-selection.com

and please visit our web site at www.natural-selection.com.

† Grant Number DAMD17-96-1-6312, the content of the information presented in this release does not necessarily reflect the position or the policy of the government, and no official endorsement should be inferred.