Diffusing into the Future

Of all of the recent functional MR techniques, there is no question that diffusion-weighted imaging has proved to be the most important. Whereas early methods were marred by technological difficulties, the advent of echo-planar imaging and improved gradients led diffusion into the spotlight. Of course, its use in the evaluation of acute infarct captured most of the early attention and still remains the most important clinical application of diffusion imaging. Other applications, however, proved the utility and robustness of the technique. These included differentiation of acute from chronic ischemia in patients with diffuse white matter changes, differentiation of cystic tumor from abscess, and differentiation of epidermoid from arachnoid cyst.

At the same time, clinical diffusion imaging has proved to be somewhat unwieldy for a number of reasons. First, due to artifacts such as T2 shine through, evaluation of lesions with diffusion often required postprocessing to create apparent diffusion coefficient (ADC) maps to eliminate the T2 component. Second, even with ADC maps, large categories of lesions (such as tumor, demyelinating disease, and infection) were found to display heterogeneous behavior with diffusion imaging. For example, some infections would show little diffusion change, whereas others would display markedly restricted diffusion. In demyelinating and dysmyelinating disease, different diffusion behavior could be documented in a single lesion. Clearly, an overall schema for interpreting diffusion images, such as is available for acute infarction, remains to be developed for other entities.

Three articles in this issue of the AJNR suggest possible future paths for diffusion imaging. DeLano et al (page 1830) used diffusion-weighted imaging with higher b values, ranging from zero to 3500 s/mm², to establish normative references for signal intensity characteristics and ADCs of the adult brain. They found that increasing b values resulted in a progressive decrease in the ratio of gray matter to white matter signal intensity. Meyer et al (page 1821) studied patients with suspected brain infarction with b values from 1000 to 3000 s/mm². They found that increased b values did not affect the diagnosis of acute infarction substantially, but did result in a marked improvement in the detection of lesions with facilitated diffusion. Clearly, findings such as these are essential to bear in mind as we progress to the use of higher b values. Finally, Melhem et al (page 1813) examined quantitative ADCs and diffusion anisotropy brain maps using six different b values, from 0 to 800 s/mm², and found that the number and strength of the b values do influence measures of diffusion and anisotropy.

Some of these articles suggest a more sophisticated use of diffusion in the future. DeLano et al's finding of variability in the gray matter to white matter signal intensity ratios has deeper implications. Unlike for field strength, where image contrast remains the same for most commonly used magnets, increasing b values have a different effect, adding another layer of complexity to our study of diffusion-weighted imaging. On the other hand, both of the articles that explore the use of higher b values also promise to simplify clinical diffusion imaging. They intimate that using increased b values may free up routine diffusion-weighted imaging from its most pressing problem, T2 shine through. At high b values, as was theoretically suggested years ago, the contribution of T2 weighting decreases. Therefore, the complicated postprocessing now necessary to achieve routine ADC maps may become a thing of the past, eliminating consideration of T2 shine through.

All three articles also serve to focus more attention on the actual physiological basis of restricted and facilitated diffusion. Whereas it is easy to consider diffusion as a measure of intracellular water, and to suggest that acute infarcts appear hyperintense due to increased intracellular water, we have always suspected that approach was too simplistic. In pure water, diffusion-weighted image intensity falls exponentially as a function of the b value. In brain tissue, water occupies many different environments, so the situation is more complex. At low b values, image intensity behaves like a rapidly decaying exponential, whereas at higher b values the intensity decreases at a lower decay rate. This behavior is often termed “biexponential” decay. The difficulty is that there is no single diffusion coefficient that describes the system; there are two. Relatively fast diffusion accounts for the rapid decay at low b values, and slower diffusion produces the gradual decay measured at high b values. Whereas it is clear that the fast and slow components depend on the properties of brain tissue, detailed studies in animals and humans have not found a simple interpretation of the two diffusion coefficients. Although it is tempting to assign the fast component to extracellular water and the slow component to intracellular water, current models of diffusion reject this interpretation (1, 2). It is likely that several factors are important, including partial volume averaging of blood and CSF, restrictions to diffusion at several length scales, and exchange of water between compartments with different diffusion and relaxation properties.

These articles also underscore the importance of standardizing the b values used in clinical studies. The benefits of improved diffusion contrast at high b values come with the complication of prescription-dependent measures of apparent diffusion. The ADC is conventionally derived from images taken at two different b values. Because tissues are described by fast and slow components, the results of a two-point measurement will depend on the specific b values chosen. If the lower b value is set to 0 (a T2-weighted image) and the upper value is allowed to vary, then the ADC will vary as a function of the upper value. Specifically, one would expect the measured ADC to decrease as the upper b value increases. This is supported by the work of Melhem et al and DeLano et al. Both articles show that the ADC varies depending on what b values are used, and that higher b values lead to lower estimates of the ADC. Melhem et al also found that higher b values (up to 800 s/mm²) reduce the standard deviations of the (isotropic) ADC and fractional anisotropy. This agrees with the results of Bito et al and Jones et al, who showed (in phantom and single subject experiments) that the standard deviations of diffusion estimates are minimized when the upper b value is approximately 1/ADC (3, 4).

Over all of this is superimposed the ultimate goal of imaging: the investigation of pathophysiology. Different b values may be better for the evaluation of different diseases. For example, demyelinating lesions may have improved conspicuity at lower b values, whereas cortical lesions may be depicted better at higher b values. Although the article by Meyer et al focuses primarily on acute infarction, it also notes that the hypointense appearance of lesions with facilitated diffusion is accentuated with increasing b values. Most striking is an example of an oligoastrocytoma, which appears isointense at b = 1000 s/mm², but markedly hypointense at b = 2000 s/mm² and b = 3000 s/mm². Clearly, much of the advantage of increased b values may lie not with the diagnosis of lesions with restricted diffusion, especially acute infarcts, but with allowing a more complete understanding of other types of disease. For example, in demyelinating and dysmyelinating diseases, the true nature of enhancing lesions may become more obvious. The differences in the diffusion characteristics of the advancing, enhancing rim versus the central portion of the lesion may be accentuated, confirming even more strongly the behavior of these types of diseases as involving not only the destruction of myelin, but also of axons in the central core of the lesion.

Ultimately, all three articles in this issue point out how simplistic much of our current approach to clinical diffusion-weighted imaging is at the moment, and how much room for future exploration remains. Diffusion imaging has become an essential part of clinical MR imaging, and it is difficult to imagine routine imaging without it. Nonetheless, we are on the threshold of an even higher level of complexity and understanding of diffusion-weighted imaging.