Normal tissue is quite orderly. Cancer, however, is "just a jumble," says Alex Hartov, an engineer at Thayer and the EIS project leader. "A lot of membranes, a lot of vascularity . . . all these things are associated with different electrical properties."

The fourth modality, near-infrared spectral tomography (NIR), uses "a very unique spectral window," says Brian Pogue, a Thayer engineer and the NIR project leader. The near-infrared wavelength range "is where the blood absorption drops way, way down, and the water absorption hasn't really increased very much," he explains. This allows electromagnetic waves to penetrate farther into the tissue and reveal information about its hemoglobin and oxygen saturation levels, both of which can signal the presence of a tumor.

Although the four modalities tackle the problem of breast imaging from different angles, several challenges tie them together. The first and most daunting one is computational complexity. Mammography uses x-rays to generate a picture of the breast. Because of their frequency and wavelength, x-rays penetrate the body in more or less a straight path. So constructing an image from x-ray data is a linear problem—a relatively easy mathematical equation to solve.

MIS, EIS, NIR, and MRE, however, require much more complicated computations. The electromagnetic waves used inMIS, EIS, and NIR may travel smoothly through some regions of the breast but be deflected, distorted, or absorbed by other regions, depending on the tissue properties in each region. Likewise, the mechanical waves generated during anMRE exam travel through breast tissue in all sorts of complex patterns.

In order to create a picture from the data gathered by each modality, the engineers needed to design software that uses what's called an iterative approach. In each modality, the signals can be

The Electromagnetic Spectrum

Depending on a woman's age, family history, and other risk factors, as well as the skill of her radiologist, the chance of a false positive ranges from less than 1% to 98% on a first mammogram. After nine mammograms, it's estimated that 43% of women will have experienced a false alarm.

measured as they are sent and received. But what happens to the signals in between, as they travel through the breast tissue, is unknown.

"You know what you're getting out, [and] you know what you're putting in," explains Margaret Fanning, an engineer who has been working on the MIS project since it began. But the researchers don't know what's happening to the signal within the

tissue. "So you guess," Fanning says, "and then you compare." Or, more precisely, the software guesses and then compares. First it guesses the spatial distribution of the tissue's physical properties; then it calculates the response that would be observed given that estimate; it compares those results to the actual data; then it makes another estimate based on the new information. The software keeps refining its guesses until the real response and the calculated response converge.

The mathematical and engineering problems involved in getting these modalities to work are "huge," says Paul Meaney, the Thayer engineer who heads up the MIS project. Several groups worldwide have tried to develop microwave-imaging techniques, for example, but "then they really fall down," he says, in trying to build an actual imaging system. In contrast, the Thayer group has managed to develop both the software and the hardware—free-standing machines forMIS, EIS, and NIR, plus specialized equipment that's used inside an MRI scanner for MRE and NIR.

But just clearing the mathematical and engineering hurdles still doesn't get the modalities to a patient's bedside. "Once [you've] got an algorithm that works, a piece of hardware that works," explains Meaney, "you start taking these images and you start to say, 'What do these images mean?'"

In other words, if the modalities are to have any future beyond the investigational stage, the engineers have to work closely with medical specialists. There are two Dartmouth clinicians who have been involved with the projects from the beginning—Poplack and pathologistWendyWells.More recent additions to the team include surgeon Richard Barth, oncologists Gary Schwartz and Peter Kaufman, and radiologist Roberta diFlorio-Alexander.

"The fun part and the hard part are that we speak different languages," says Poplack. The benefit, he adds, lies in the