COLUMBUS, Ohio -- Doctors may one day be able to diagnose breast cancer with better accuracy using a new imaging method being developed at Ohio State University.
The method, which involves computerized analysis of magnetic resonance images (MRIs) of breasts, could greatly reduce the number of women who have to undergo painful biopsies. With current screening techniques, seven out of every 10 women suspected of having breast cancer and sent for biopsies turn out to have no malignancies.
To lower that high ratio of false positives, a team of Ohio State researchers, led by Bradley Clymer, an associate professor of electrical engineering, is developing a diagnostic step between X-Ray mammography - the technique most commonly used for breast cancer screening - and biopsy. In this step, doctors would analyze MRIs of breasts to detect blurred regions which might correspond to microcalcifications that are often indicators of cancer.
In a recent study, their method - which is still several years away from clinical use - was nearly 100 percent accurate in detecting blurred regions in breast MRIs. The study was published in a recent issue of the Journal of Magnetic Resonance Imaging.
The method is based on the knowledge that calcified regions in the breast show up as blurred aread in MRIs - a result of the magnetic properties of calcifications being slightly different from those of normal soft tissue. Some of these calcifications, which occur when breast tissue hardens due to calcium accumulation, are a sign of malignancy.
Working with Petra Schmalbrock, an associate professor of radiology, and graduate student David James, Clymer showed that it is possible to detect blurred patches in MRIs using a computer program that recognizes minor differences between pixels. The program, developed by Clymer's group, can detect blurs invisible to the naked eye.
Clymer is now studying how specific kinds of calcification in the breast - like a cluster of tiny particles or branching lines of hardened tissue - would distinctively blur a breast image. "Once we have a vocabulary of blurred signatures for different types of structures in different orientations and locations in the breast, we can link them to the probability of cancer in each of those regions," Clymer said.
MRI data could then be used to confirm or reject suspicious regions found on X-Ray mammograms. "This would hopefully reduce the number of biopsies in women with no malignancies," Clymer said.
Although the MRI test could end up being slightly more expensive than a biopsy, Clymer thinks it would still be attractive because of its non-invasive nature.
To test the feasibility of detecting blurs on MRIs, Clymer and his colleagues took MRI images of healthy breasts and introduced blurred patches in them. This they did by blurring an image completely on a computer and pasting small parts from this blurred image on specific locations in a normal image. The blurs were superimposed in a way that simulated a certain type of calcification known as "focal clusters".
For the detection, the researchers used statistical texture analysis - a method to observe how pixels (individual picture elements) are related to their neighbors within an image. "For example, in an unblurred image, we would expect sharp differences between neighboring pixels at the borders between different types of tissues," Clymer said. "Conversely, in a blurred image, the image has a more gradual variation in the gray tones from one pixel to the next."
The software that the researchers used produced 14 measures of difference between pixels - called texture features, quantifying the extent of blurring in different locations of the image. "Combining these measures, we could tell whether the image was blurred within a certain region and to what degree," Clymer said.
The researchers found that the software was able to detect more than 90 percent of the blurred regions. The accuracy of detection was close to 100 percent. "Using the method, there would be very little likelihood of incorrectly detecting regions where there is no blurring," Clymer said.
Clymer's group is now working to model the blurring effect of specific types and geometries of calcification. "Then we can distort a normal image in a particular way - corresponding to a specific type of tumor - and check if the distortions can be detected accurately by statistical texture analysis," Clymer said. "If we can show that it works for all possible kinds of calcification, we will be ready to study human subjects."
In two years, the researchers hope to be able to work with radiologists to test the method for actual diagnosis. "We would get women at the screening stage (undergoing X-Ray mammography) to also have MRIs taken," Clymer said. "We would run the images through our software and see how well our diagnosis matches biopsy results."
The above post is reprinted from materials provided by Ohio State University. Note: Materials may be edited for content and length.
Cite This Page: