MRI versus Mammography

MRI has certain advantages that make it attractive for the evaluation of breast disease: Virtually all malignancies enhance after contrast injection, MRI has a high sensitivity for the detection of breast cancer--95% or greater, according to several studies ^1,2 --and it is able to visualize very small tumors (<5 mm). Defining precisely how small a tumor MRI
can visualize has been challenging. It is possible to see submillimeter enhancement that comprises only a few pixels, but this is a fleeting effect that is impossible to validate pathologically.

In addition, MRI is not limited by breast density, which is a major advantage in comparison with mammography, and is effective even in women who have breast implants or have undergone previous breast surgery. Finally, MRI has a very high negative predictive value. Therefore, when no enhancement is seen after a successful injection of contrast, it is very unlikely that malignant disease is present.

One of the disadvantages of breast MRI is that its specificity appears to be moderate at best, although the specificity has been difficult to define. Reports in the scientific literature cite values of 37% to 97%, ^1,3-11 but these findings appear to depend heavily on the selection of both the MRI technique and the criteria for determining malignancy.

Additional disadvantages of breast MRI are its relatively long scan times (multiple sets of images are usually acquired during a 30- to 45-minute examination) and its requirement for contrast injection, which is unnecessary in mammography. The high cost of breast MRI is also a disadvantage.

Although mammography and breast MRI are different in many ways, they are in fact complementary procedures and work well in combination. Mammography, for example, readily detects calcifications, whereas MRI does not. Mammography loses some of its effectiveness in dense breast tissue, whereas MRI appears to work better in dense breast tissue than in fatty tissue, because fewer fat-water boundaries mean a better, more homogeneous signal.

Breast MRI Technique

Contrast is considered essential for the evaluation of breast cancer. A single dose of gadolinium contrast media, 0.1 mmol/kg body weight, is used in most breast MRI studies. Early studies with spin-echo techniques used double-dose contrast, but today most examinations involve gradient-echo sequences in which a single dose of contrast is most effective. Precision isn't critical in contrast administration. Injections need to be consistent but do not necessarily require the delivery of a fast bolus. Manual injection is acceptable.

A dedicated breast coil is required. Patients generally lie in the prone position in order to minimize the image blurring that can be caused by cardiac and respiratory motion. Some devices have compression plates for holding the breast stable, but in most cases, with a little coaching, patients remain still during the scan, and image registration is not a problem. With certain coils, patients can go into the magnet feet first, which seems to reduce feelings of claustrophobia substantially.

T1-weighted imaging is required for the best sensitivity to the gadolinium contrast agent; however, the selection of other imaging parameters depends on the clinical question at hand, the available equipment, and the radiologist's preference. Variables include slice thickness and orientation (transaxial, coronal, or sagittal); the tradeoffs between temporal and spatial resolution; and choices between unilateral or bilateral acquisition, 2- or 3-dimensional imaging, spin-echo or gradient-echo sequence, and injection of contrast by hand or by power injector.

Morphology or Function?

The strategies used in breast MRI depend on whether the goal is diagnosis or staging. In staging, it is important to determine the size and extent of a cancerous lesion and whether it is multifocal. The priority is the sensitivity of imaging. Three-dimensional spoiled gradient echo techniques are generally used (Table 1) because of their high spatial resolution, ability to cover the full volume of the breast, and inclusion of fat suppression. Scans are longer when the goal is staging, usually taking 2 to 4 minutes to ensure both adequate coverage and a high signal-to-noise ratio, a measurement of image quality.

When the purpose of MR is to diagnose a suspicious lesion, specificity is the priority. In this case, dynamic techniques using 2D spin echo sequences are generally used (Table 2); however, 3D techniques can be used with reduced resolution, and scan times will be similar to those of dynamic 2D methods. Dynamic scans, used to observe contrast uptake, usually take 1 minute or less. In fact, 15-second scans are not uncommon. To achieve high temporal resolution, it is often necessary to limit coverage, reduce spatial resolution, and/or forego fat suppression. From time-intensity curves, pharmacokinetic parameters related to vessel permeability and blood volume can be calculated.

The approaches to high-resolution and dynamic scans are beginning to converge. Faster 3D sequences equivalent in speed to 2D spin-echo techniques are possible and, as a result, dynamic scanning is increasingly using 3D methods. In general, the 3D methods are not extremely high in either temporal or spatial resolution; however, they balance the need for both enhancement morphology and enhancement kinetics. For now, most practitioners prefer image quality to be high enough to identify specific features that raise diagnostic confidence--for example, the interior and border of lesions. Once image quality is adequate, temporal resolution can be increased as much as possible.

Finally, the limited availability of computer-aided kinetic analyses also influences the preference for studies that emphasize morphology over function. There is a general consensus that MRI should be integrated into breast imaging practices, and that breast radiologists should do the interpretations. However, without the aid of effective computer tools to process data rapidly, the interpretation of functional information and generation of parametric images is too time-consuming for most mammography practices.

Figure 1 shows a 3D contrast-enhanced breast MR image acquired with high spatial resolution in a woman with invasive carcinoma in the lower half of the breast. The lesion is very well characterized morphologically, and spiculations and dark areas within the lesion are clearly evident.

Figure 2 depicts the same lesion, in this case acquired using a 2D dynamic MR sequence, a 15-second scan time, and no fat suppression. The lesion is visible, but the internal morphology is not as apparent as on the 3D image. It is possible on the dynamic image, however, to observe the time course of tumor enhancement. Contrast washout after 2 minutes, or a fast increase with either stabilization or washout of contrast, as seen in Figure 2, are both considered suspicious findings. A more gradual uptake without washout is considered less suspicious. The specificity of these patterns is a problem, however. The gradual pattern of enhancement could also be consistent with ductal carcinoma in situ (DCIS). In cases of DCIS, a linear or segmental morphologic pattern is extremely helpful in making a diagnosis.

Whether to do bilateral or unilateral imaging is also a source of ongoing debate. The argument for unilateral imaging is that keeping spatial resolution as high as possible enables careful evaluation of symptomatic lesions, particularly their morphology.

There is also a strong argument for the importance of symmetry, which can be evaluated only with bilateral imaging. For example, a regional area of enhancement is considered less suspicious for malignancy if it is observed in both breasts. When referring a patient for surgical staging following detection of a symptomatic breast lesion, physicians increasingly request evaluation of the other breast. This represents a screening application of breast MRI--however, one that is the subject of going research but is not yet ready for clinical use.

Emerging Applications

Differential diagnosis represents an emerging application of breast MRI. Indications that fit in this category include an inconclusive mammogram, a palpable abnormality without mammographic findings, and nipple discharge without mammographic findings. MRI is also used to differentiate recurrent breast cancer from scar tissue.

For these indications, MRI has a high sensitivity, but only a moderate specificity. The frequency of its use for differential diagnosis depends on how heavily minimally invasive biopsy procedures, including fine-needle aspiration, stereotactic core biopsy, and ultrasound-guided core biopsies, are used in a particular practice. In many areas of the United States, where these invasive procedures are common, there is less demand for MRI to determine whether a lesion is malignant. In Europe, however, breast MRI is more widely used for differential diagnosis.

The patient in Figure 3 had a spiculated mass on mammography, which may have been either recurrent carcinoma or scar. On MRI, a distortion caused by a previous biopsy is apparent, but there is no contrast enhancement. Incidental enhancing foci found elsewhere in the breast were determined to be ductal carcinoma in situ.

Staging is another emerging application of breast MRI. It is useful for staging in the case of a biopsy that shows cancer, tissue margins that test positive for cancer cells following lumpectomy, or a cancerous lymph node whose source lesion has not been identified in the breast. There is an increasing need to stage the extent of disease preoperatively, as needle biopsy procedures provide no surgical margins for pathological examination, as excisional biopsy does.

Assessment of the response to preoperative chemotherapy is an additional staging application. In this indication MRI can be used to: 1) stage the extent of disease to determine the appropriateness of preoperative chemotherapy; 2) monitor the response during treatment and, potentially, change the course of treatment for patients with non-responsive tumors; and 3) measure the extent of disease following chemotherapy to determine if breast conserving surgery is an option.

In staging applications, MRI is proving to be more effective than mammography. There are several reasons for this. Contrast-enhanced MRI is very sensitive to breast carcinoma. Its 3D format results in a superior anatomical representation. Breast MRI is more accurate for demonstrating tumor extent. Both mammography and MRI demonstrate malignancy; however, concordance with the pathological determination of the extent of disease was found in one study to be considerably higher for MRI: 98%, as compared with 55% for mammography ¹² (Table 3).

MRI offers a particular advantage when DCIS is present (Table 4). As shown in Figure 4, MRI easily depicts multifocal disease, as well as significant axillary involvement, and the distribution of DCIS over an entire segment of the breast.

Screening represents the most challenging application of breast MRI, but it may offer the greatest potential to improve patient outcomes. Genetic testing and statistical models can identify women at high-risk for breast cancer, but early detection has been more difficult. High-risk women are likely to develop cancer at a young age, when breast tissue is still very dense, and mammography is less effective. Until recently, one of the few options for reducing breast cancer risk has been bilateral prophylactic mastectomy.

MRI may offer an effective alternative for screening and surveillance. It is not considered practical in a general population, but MRI screening is proving useful in women who have dense breasts or have been determined to be at high risk--those who carry BRCA1 or BRCA2 gene mutations, have a personal or family history of breast cancer, or have cancer in the contralateral breast.

MRI does have shortcomings when used for screening, however. Incidental enhancing lesions are identified in many patients. Most will turn out to be benign, but positive findings on MRI create a high level of anxiety. Biopsy is difficult, as the lesions are neither palpable nor seen on mammography, and MR-guided biopsy tools are insufficiently developed.

Several trials of MRI for breast cancer screening are under way in the United States, Canada, and Europe. Participants are BRCA1 or BRCA2 mutation carriers, or have a high risk for breast cancer on the basis of family history or statistical modeling. They undergo an initial screening MRI, with follow-up examinations at least annually. Data from these studies will be pooled for the evaluation of sensitivity and specificity. Early results suggest that MRI screening detects breast cancer in 2% to 3% of high-risk women, a reasonably high rate. ^13-15

Future Developments

As effective computer-aided analyses become widely available, a greater emphasis may be placed on fast imaging. The specificity of MRI could be improved by the introduction of better contrast agents that yield more accurate pharmacokinetic measurements or that target cancer cells specifically. These agents would likely reduce lesion conspicuity, however, so their overall benefit is not clear.

Improvements are clearly needed in localization and biopsy methods. This might involve localization of lesions for subsequent ultrasound-guided, stereotactic, or excisional biopsy, or the development of MR-guided biopsy tools. Finally, there is intense interest in the development of tools for noninvasive tumor ablation by MR-guided cryotherapy, radiofrequency energy, or focused ultrasound heating. *

Discussion

TG: Thank you very much, Dr. Hylton. You mentioned that specificity is still an issue. There have been some recent reports using dynamic susceptibility imaging, following the initial T1-weighted gradient echo. So, this could be analogous to what Dr. Rowley showed in the brain, where altered perfusion could be identified, probably reflecting angiogenesis in the lesion.

That would have an impact on the most desirable contrast agent, because you may want to give that at a high rate of injection. What do you think about that in the future?

NOLA HYLTON, MD: I think that it's likely that in the future that we'll do something like that. Once the ability is there to augment the morphologic information with a functional over-lay, then perfusion makes sense, and actually Christiane Kuhl, in Germany, published some work comparing T2 perfusion imaging in breast lesions. I think her specificity was either equivalent or slightly better than dynamic T1-weighted contrast evaluation. These are types of things where the diffusion information would be added.

But as I mentioned earlier, I believe it could be useful as long as it doesn't compromise what you can see morphologically about the lesion. There are important features that have become associated with certain types of important histologies, for example ductal carcinoma in situ has some patterns that requires high resolution in order to be identified on MR. A lot of the ability to diagnosis depends on what we see at the borders and interior of enhancing lesions. So, as long as those features are maintained, this is an early detection method. Clearly, we also want to be able to characterize small lesions, because we hope to use this in a screening capacity, so we will need to see these lesions and characterize them when they're still fairly small. So the addition of techniques that add physiologic or functional information would likely be done with a second bolus, and would likely be used to overlay functional information on what we can already see anatomically.

TG: But it would be in a second bolus situation. So you would do your T1-weighted gradient echo and then the second bolus with susceptibility imaging or some other function.

NH: Well, there are a few other clever pulse sequences that people are developing to combine both high spatial resolution and high temporal resolution. I believe it came out of your own work and you are familiar with it. But, in terms of taking cores or samples of k-space, you need to get the high temporal resolution and information about contrast enhancement, then combine it to create one higher resolution 3D, longer time scan, imaged to see the morphology.

DR: I've been working with a method based on a similar idea to TRICKS, but using a different geometry of acquisition in the k-space. It offers surprisingly good spatial resolution and temporal resolution, sort of an optimal solution to that problem.

NH: I think something like that is very attractive. As long as the reconstruction tools, etc., are there to apply these techniques, we'll get the best of both worlds out of the data.

DB: Ultimately, you need to cover both breasts very rapidly with high spatial resolution. How fast do you think we need to image, what temporal resolution behind which is no longer worthwhile? As you go to faster acquisitions, do the injection rates start to matter more and have to be controlled more?

NH: My personal belief is that anything below 30 seconds is going to compromise imaging, compromise the spatial resolution. But I am very influenced by the argument that we do need to be imaging bilaterally, which is actually going against getting better spatial resolution. So if we realize any improvements in imaging, it would probably be to get more effective bilateral techniques. I don't see in the near future that we'll be going incredibly fast in scanning, unless we're doing two different types of techniques, and looking at them in combination.

For the implications for contrast and contrast injection, I would only speculate that if you do some sort of sequential type of an approach, the secondary functional information would be very dependent on the injection dose, mode of administration, and rate, etc., and that those things would need to be very controlled, likely would be best done with a power injector. The practicality of the use of a power injector is that if you're not there for every exam, at least you have some confidence that it was done in a consistent way. So, you are not asking yourself on any given study, if there could have been a problem with the injection.

MP: There's an emerging concept in SMASH, in so-called parallel imaging, which can help to solve the trade-off of being able to image fast and simultaneously at high resolution, but introducing the additional trade-off of sacrificing signal-to-noise. One way of buying back signal-to-noise is to increase your dose of gadolinium. Right now since you are only operating at a single dose of contrast, you have a lot of room to increase the dose.

What do you think about this possibility that with techniques that sacrifice signal-to-noise to address the temporal and spatial resolution issues, that you'd be willing to buy that back by bumping your dose up?

NH: You know, at the moment, we're hitting an over-sensitivity with respect to the objects that enhance. I think by increasing the dose, we tend to see more of the hyperplasias of fibrocystic disease, or proliferative diseases that tend to light up, and that we have a lot of problems trying to decipher them. A lot of the attempts to differentiate hyperplasias from DCIS have to do with seeing if you can see changes in the very initial uptake. If you pump up the contrast dose, you might wash some of that out, you might lose some of that. I'm not sure that dose as a method of getting better signal-to-noise is necessarily what the issue is.

I think that the signal-to-noise needs simply to allow us to see anatomically what's there. But I think if you use contrast to get that effect, you are going to lose information about border differentiation, the rate of enhancement over time, and whether you have a rim versus a central enhancement, etc. Those are things that we actually look at and make decisions about.