Adding Noise Shown to Improve Mammogram Accuracy

It does this by adding carefully selected noise to the system.

The result has been a distinct improvement in the system's ability to correctly identify precancerous lesions, plus a 36 percent reduction in false-positives. The inventors have developed an innovative method of calculating precisely the correct type and level of noise to add to existing noise in radiography or a similar system. "We see a broad spectrum of applications for this technology,” stated research assistant Prof. Hao Chen, from Syracuse University (SU; NY USA). "If a system's performance is unsatisfactory, we add noise to the system based on a specific algorithm that can significantly improve system performance.”

A patent covering the technology has been issued to Prof. Chen, Prof. Pramod K. Varshney and research professor James Michels. All are associated with SU's L.C. Smith College of Engineering and Computer Science.

In mammography studies conducted by doctoral candidate Renbin Peng, the challenge was to identify clusters of microcalcifications in breast tissue. These early signs of precancerous conditions average only 0.3 mm in size and offer only slight contrast with surrounding tissue. In addition to improving detection of these lesions, the researchers have reduced false positives by more than one-third.

While the current focus of the research group is on medical uses of stochastic resonance, other applications are expected in enhancing audio, video, geophysical, environmental, radar, and other signals. The group has been receiving support from the U.S. Air Force Office of Scientific Research. Ongoing investigations by the Syracuse group are expected to produce further improvements in the efficiency and robustness of the SR-based detection techniques.

An article by the inventors on the theory of stochastic resonance effect in signal detection was published by IEEE Transactions in Signal Processing in July 2007. Another article, covering the mammography studies, coauthored by Mr. Peng, was published in IEEE Journal of Selected Topics in Signal Processing in February 2009.

Related Links:
Syracuse University