| Volume 4, | Number 1 | March/April 1996 |
A unique marriage between Hubble Space Telescope astronomers and cancer researchers has produced an image-processing technique that shows promise in detecting early breast cancer. Employing techniques used to correct the blurry images sent by Hubble prior to the 1993 servicing mission, this method is designed to detect microcalcifications, an early sign of breast cancer. A group of astronomical and medical researchers from the Space Telescope Science Institute (STScI) in Baltimore, Johns Hopkins University, and the Lombardi Cancer Research Center at the Georgetown University Medical Center in Washington, D.C., is testing this technique to detect microcalcifications in digitized mammograms.
Detecting a microcalcification among the background structures in a mammogram is remarkably similar to finding a faint star in a blurry and cluttered telescope image. Dr. Benjamin Snavely, the National Science Foundation's (NSF) program director for advanced technologies and instrumentation in astronomical sciences, noticed that certain medical images of interest to the Lombardi Center were similar to the astronomical images STScI scientists obtained from Hubble.
Dr. Snavely saw promise in teaming the disciplines, and he arranged for a meeting between the Lombardi Center and STScI. "Each group immediately became interested with what the other was doing," said Dr. Snavely. "They struck up a resonance." The collaboration was awarded a $50,000 grant from the NSF.
Using Hubble's image-processing lessons for cancer detection is the silver lining to the story of Hubble's spherical aberration. When the now-corrected flaw in Hubble's primary mirror was discovered, STScI developed a large repertoire of image-processing software to correct for the telescope's loss of dynamic range and spatial resolution. The flaw allowed STScI scientists to become experts in image-processing techniques that they otherwise would not have needed. STScI is funded by NASA to study the data received from Hubble.
STScI's Dr. Robert Hanisch and Dr. Richard White used a three-part image-processing technique to identify calcifications in four separate test cases, two of which were blind. The key step involved variance normalization, a technique the astronomers had advanced much farther than the medical researchers.
The team's techniques should lead to a more unbiased detection of smaller lesions. The conventional method for detecting lesions is through "eyeball" inspection, which carries the risk of human error.
Next, the team plans to run tests against a standard set of digitized mammograms, measuring overall performance in comparison to other methods of microcalcification detection. These tests will also allow the team to determine the size and sensitivity limits necessary for reliable identifications.
| Potential Applications of Astronomical Image-Processing Techniques to Digitized Mammograms |
|---|
| Shown are four versions of a digitized mammogram that was processed using methods developed for astronomical images. The image at upper left is an enlargement of the region of the original mammogram with suspected micro-calcifications. The image at upper right is the result of a standard unsharp mask image-processing operation. This operation enhances small-scale structures, such as the microcalcifications we are seeking, but also enhances many other structures so that the microcalcifications remain difficult to detect. The image at lower left shows a modified unsharp mask, which uses a technique of variance stabilization to assure that small-scale structures are detected uniformly throughout the image, independent of the local intensity level. Finally, the image at lower right shows the result of applying an adaptive filter to the image at lower left. This filter smoothes out the structures that are statistically insignificant, leaving just the structures that are most likely to be identified as micro-calcifications. This method has greatly enhanced the visibility of bona fide microcalcifications that had been seen by direct inspection of the original x-ray images, but might have easily escaped detection had they been smaller or located in a more confusing region of the image. The approach should make it possible to improve the reliability of early diagnosis of breast cancer potential cases. |
For more information about the technique, contact Mark Jaster at the Goddard Space Flight Center. Phone: 301/286-9232, E-Mail Mark.Jaster@ccmail.gsfc.nasa.gov Please mention that you read about it in Innovation.
