Diffusion tensor imaging: Basic principles and emerging clinical applications


View content online at: http://www.appliedradiology.com/Issues/2003/06/Supplements/Diffusion-tensor-imaging--Basic-principles-and-emerging-clinical-applications.aspx

Abstract:  Diffusion tensor imaging (DTI) is a rapidly growing area of interest in radiology research. By querying tissue at the microscopic scale, DTI provides researchers and clinicians with new biological information not previously available with conventional magnetic resonance imaging. Basic concepts of water diffusion and DTI are reviewed in this manuscript, followed by a brief review of topic areas of ongoing clinical research and developing clinical applications, including brain development, brain injury, white matter diseases, epilepsy, neuropsychiatric diseases, and brain tumors.
Loading...

Dr. Maas is currently a third-year Resident in Diagnostic Radiology at the University of California, San Francisco. He earned his SB, SM, and PhD degrees from the Massachusetts Institute of Technology and his MD degree from Harvard Medical School, Cambridge, MA.

Diffusion tensor imaging (DTI) is a rapidly growing area of interest in radiology research. By querying tissue at the microscopic scale, DTI provides researchers and clinicians with new biological information not previously available with conventional magnetic resonance imaging. Basic concepts of water diffusion and DTI are reviewed in this manuscript, followed by a brief review of topic areas of ongoing clinical research and developing clinical applications, including brain development, brain injury, white matter diseases, epilepsy, neuropsychiatric diseases, and brain tumors.

Diffusion tensor imaging (DTI) is an area of rapidly growing research in both technical refinements and clinical applications (Figure 1). Diffusion data reflect information on a microscopic scale, allowing researchers an unprecedented tool to noninvasively probe tissue microarchitecture in vivo. This manuscript aims to provide readers with an introduction to the basic concepts of water diffusion and diffusion tensor imaging, followed by a brief review of areas of ongoing clinical research and developing clinical applications, with emphasis on areas of research presented at the 88th Scientific Assembly and Annual Meeting of the Radiological Society of North America (RNSA 2002).

Physical basis of diffusion

Molecules are constantly undergoing random motion as a result of their thermal energy. The Einstein equation states that the mean square molecular displacement along a given axis is equal to twice the product of the observation time and a constant, D, known as the diffusion coefficient, measured in mm 2 /sec. For the typical echo times of 50 to 100 msec and diffusion coefficients encountered in DTI, root mean square diffusion of water is on the scale of 10 to 15 µm. Thus, DTI can provide information related to tissue microarchitecture at this scale, as diffusing water molecules probe cell size, shape, and organization during each acquisition.

If molecules are free to diffuse in all directions with equal probability, as in pure water, diffusion is described as isotropic and a single diffusion coefficient, D, is sufficient to describe the expected diffusion behavior. However, in biological tissues, the mobility of molecules is affected by the local cellular microarchitecture. When the obstacles to mobility are different depending on the direction a molecule travels, diffusion is described as anisotropic, and a different diffusion coefficient may be measured along each direction.

An example of a setting with anisotropic water diffusion is a white matter tract, 1 where the tight packing and parallel orientation of axon membranes and the presence of myelin sheaths hinder diffusion in directions perpendicular to the axons, such that diffusion parallel to fiber tracts occurs "faster" than diffusion perpendicular to the tracts. Anisotropy has been observed in many other tissues, including skeletal muscle 2,3 and myocardium, 4 as well as cortical gray matter of premature infants. 5

The diffusion tensor

The diffusion tensor formalism 6 was first applied in multidimensional magnetic resonance imaging (MRI) by Basser et al 7,8 in order to simplify analysis of anisotropic diffusion data. In this model, diffusion at each voxel is described by a symmetric 3 * 3 matrix of diffusivities, D , known as the diffusion tensor, where the bold face is used to indicate that D is a matrix. The diagonal elements of this matrix, Dxx, Dyy, and Dzz, correspond to the diffusion coefficients estimated with gradients applied along the principal axes of the reference frame, ie, the scanner frame. The off-diagonal elements, Dxy, Dyx, Dxz, Dzx, Dyz, and Dzy, represent correlations between random motions along the respective paired axes of the reference frame. The matrix is symmetric, ie, Dxy = Dyx, Dxz = Dzx, and Dyz = Dzy.

In anisotropic diffusion, diffusion is directionally dependent and correlations in diffusion between the axes of the reference frame are possible, such that the tensor may contain all nonzero values. In diffusion tensor analysis, the tensor is diagonalized, ie, rotated mathematically onto another set of axes that eliminate the off-diagonal correlations, known as the principal axes of diffusion. Once this has been accomplished, the diagonal elements of the rotated matrix, known as the eigenvalues, represent the diffusion coefficients along the principal axes formed by the corresponding eigenvectors. The largest eigenvalue is known as the major eigenvalue, and the corresponding major eigenvector represents the "preferred" direction of diffusion at each voxel.

The eigenvalues and eigenvectors can be graphically represented by an ellipsoid (Figure 2). The major axis of the ellipsoid is parallel to the major eigenvector and the minor axes parallel to the minor eigenvectors, with the size of the ellipsoid along each axis related to the corresponding eigenvalue. The eccentricity of the ellipsoid thus graphically reflects the anisotropy of diffusion. In the case of isotropic diffusion, the ellipsoid reduces to a sphere.

Measuring diffusion with MRI

As a proton diffuses across a magnetic field gradient, irreversible spin dephasing occurs, reducing the measured MR signal. 6,9 By analyzing the degree of signal loss observed after application of diffusion-sensitizing gradients, also known as motion-probing gradients, the underlying diffusion coefficient can be estimated. To simplify analysis, the shape, strength, and duration of diffusion-sensitizing gradients are summarized by the b factor, measured in seconds per millimeter squared. For typically encountered b values of <1500 sec/mm 2 , attenuation is well approximated by a monoexponential model. Thus, the diffusion coefficient D i along a given direction i is then estimated by solving the equation I = I(0)exp(-b i D i ), where I is the observed signal after application of the diffusion-sensitizing gradient characterized by b i and I(0) is the signal without application of the gradient, ie, b = 0.

In isotropic diffusion, D is independent of direction and thus the choice of direction of the diffusion-sensitizing gradient is arbitrary. With anisotropic diffusion, however, the observed signal varies with different gradient directions, and a maximal signal loss is observed when the gradient is aligned with the major axis of diffusion. In practice, however, the principal directions of diffusion are unknown or spatially varying, and full characterization of diffusion characteristics requires estimation of the complete diffusion tensor. In addition to the unweighted image (b = 0), measurements with at least six noncollinear gradients are required to estimate the six independent components of the matrix, for a total of at least seven acquisitions comprising a typical complete DTI dataset. From the imaging data, the diffusion tensor is then filled by least-squares approximation of its independent components and then its eigenvalues and eigenvectors solved. Additional measurements decrease susceptibility to noise, with additional gradient directions at a single b value providing better signal-to-noise ratio than an increased number of averages with fewer directions or an increased number of b values. 10

Measurements derived from the diffusion tensor

Once the tensor has been estimated, many measures characterizing different properties of diffusion can be calculated for each voxel. Various scalar measures relating to average diffusivity and the degree of diffusion anisotropy are described below, as well as methods to express the three-dimensional information included in the diffusion tensor related to the direction of maximal diffusivity.

Diffusivity

The mean diffusivity is equal to the trace of the diffusion tensor, ie, the sum of the diagonal elements, divided by 3. The trace is a rotationally invariant measure, ie, it is independent of the reference frame used, and is thus always equal to the average of the three eigenvalues of the diffusion tensor, ie, the mean of the diffusivities along the three principal axes of diffusion. The mean diffusivity is often referred to as the apparent diffusion coefficient (ADC), where the use of the word apparent acknowledges the effects of tissue interactions as opposed to a true free diffusion coefficient.

Examples of normal measured ADC values (*10 -3 mm 2 /sec) are 3.2 in cerebrospinal fluid (CSF), 0.83 in adult cortical gray matter, 0.65 in adult white matter, and 1.1 to 1.6 in neonatal brain tissue.

Anisotropy

Measures to quantify the degree of anisotropy are also desirable. Commonly used measures of anisotropy derived from the eigenvalues of the diffusion tensor include fractional anisotropy (FA), 11 relative anisotropy (RA), 11 normalized relative anisotropy (A*), 12 and the volume ratio (VR). 13 Values of FA vary from 0 (isotropic) to 1 (infinite anisotropy). Similarly, RA varies from 0 to the square root of 2, (A*) from 0 to 1, and VR from 1 to 0. Volume ratio is often expressed as 1-VR, so that isotropy is represented by a value of 0. Measures to evaluate the coherence of anisotropic diffusion between a voxel and its neighbors, such as the lattice index (LI), 13,14 have also been developed.

Of these measures, FA is most commonly used, as it has the most favorable signal-to-noise characteristics. 15,16 A map of fractional anisotropy in the brain of an adult volunteer is presented in Figure 3.

Directionally encoded color maps

The eigenvector of the diffusion tensor corresponding to the largest eigenvalue is known as the major eigenvector and represents the preferred direction of diffusion. In a white matter tract, for example, this direction corresponds to the direction of the fiber tract. Development of display strategies to express this three-dimensional information on two-dimensional monitors is an area of ongoing research. A number of methods have been described to use color to encode the direction of the major eigenvector. 17,18 A frequently used method is to set the red, green, and blue values at a pixel according to the right-to-left, anterior-to-posterior, and superior-to-inferior components of the major eigenvector, with the overall intensity modulated by the fractional anisotropy. In this way, it is possible to separate adjacent white matter tracts in a manner not possible with conventional MRI, as seen in the color-encoded directional map of the brain of an adult volunteer presented in Figure 3C.

Whisker plots

Another technique of displaying directional information is known as whisker plotting. At each pixel where anisotropy is present, a short linear "whisker" is superimposed upon the image and aligned parallel to the in-plane projection of the major eigenvector, thus providing the observer with a detailed view of pixel-by-pixel directional information in the imaging plane. These plots are also known as fiber orientation plots, as each whisker indicates the dominant orientation of fibers at the corresponding position. A whisker plot of the brain of an adult volunteer is presented in Figure 4. While major tracts, such as the corpus callosum, that curve within the imaging plane are more easily visualized with whisker plots than color-encoded images, a limitation of the whisker plot method is the difficulty in displaying eigenvector information perpendicular to the plane of the image, such as corticospinal tracts in axial images.

Fiber tractography

Since the major eigenvector of the diffusion tensor in white matter voxels is assumed to be collinear with the direction of a white matter fiber tract, the directional information can be used to map tracts by connecting neighboring pixels that lie along the direction of this eigenvector. 19-21 This process, known as fiber tractography, is also an area of much active research. All fiber tracts passing through a single voxel can be examined, or, alternatively, all tracts connecting specified "start" and "end" positions can be explored. If prior knowledge of anatomic connections is incorporated, major known fiber bundles can be mapped, such as the corticospinal tracts. An example of tractography in the setting of a brain tumor is shown in Figure 5.

Brief review of applications

Normal development and perinatal brain injury

Several groups have explored the normal changes in diffusion properties of the human brain during development, as well as alterations to these patterns following neonatal brain injury. Simple models to describe white matter and deep gray matter diffusion magnitude and anisotropy from infancy to adolescence have been developed, suggesting that certain DTI milestones for brain maturity can be established. 22,23

Rapidly decreasing brain ADC has been observed during the first 2 years of life, 22,24,25 with a more gradual decrease during the next few years as brain ADC approaches adult values. 22 The basis of this decrease is incompletely understood but is generally attributed to increasing myelination and the formation of new barriers to diffusion with brain maturation, as well as decreasing extracellular free water.

Similarly, in studies of white matter, increasing anisotropy with age has been observed in preterm newborns, 26,27 infants, 22,25 and children. 22 Like ADC, the change in anisotropy is most rapid during the first 2 years of life, and more gradual thereafter. 22 Thus, diffusion anisotropy may provide a marker of normal brain white matter maturation and myelination. 28,25 For example, in a study of normal children aged 4 days to 71 months, higher anisotropy was seen in compact than in noncompact white matter at all ages, but greater increases in anisotropy were observed with age in noncompact white matter, suggesting that myelination is more advanced in compact white matter at birth, with greater changes in myelination occurring in noncompact white matter following infancy. 29 Anisotropy, however, is not only a marker for myelination, as anisotropic diffusion is also seen in nonmyelinated nerve bundles 30 and white matter prior to myelination, 25,31,32 suggesting that in addition to myelin, the dense packing and parallel alignment of axonal membranes contribute to anisotropy. Increasing fiber bundle organization may also play a role. 33

Diffusion tensor imaging may also have a role in the evaluation of maturation of cortical gray matter. In a study of neonates of 26 to 41 weeks gestational age, 5 significant anisotropy of cortical gray matter was seen in the most premature infants, with a radially aligned major diffusion tensor eigenvector, consistent with the relatively simple radial neuronal architecture at that stage of development. Anisotropy essentially disappeared by 36 weeks gestational age, correlating with the normal increase in the complexity of cortical architecture as neuronal connections are formed as a fetus approaches term.

Many studies of perinatal brain injury have also been performed. Apparent diffusion coefficient imaging may be superior to conventional imaging in evaluation of perinatal brain injury, as one study has found that abnormal decreases in ADC seen on diffusion imaging may better demonstrate and define the extent of perinatal brain injury than conventional MRI, especially when obtained between the second and fourth days of life. 34 In a study of premature infants imaged near birth and again near term, 35 infants with moderate white matter injury did not demonstrate the typical decrease in brain ADC and increase in white matter anisotropy observed in controls who were born at term. Those neonates with minimal white matter injury showed similar decreases in ADC to the control infants, but did not show the normal increase in frontal white matter anisotropy, suggesting a more mild impairment of white matter development.

In another study of perinatal brain injury, abnormally decreased anisotropy was observed at the site of the original central white matter injury as well as in the ipsilateral internal capsule when compared with controls, suggesting impaired development of these fiber tracts. 36 In a complementary study of two 6-year-old boys with spastic quadriplegia secondary to known periventricular leukomalacia (PVL) in the neonatal period, DTI demonstrated attenuation of white matter fiber tracts projecting to and from occipital and parietal lobes, with preservation of the corticospinal tracts, suggesting that pathophysiology of motor disability in PVL may be related to abnormal sensory cortex projections, rather than pyramidal motor tracts, as generally postulated. 37

Ischemic and traumatic brain injury

The most well-established clinical application of diffusion imaging is in the evaluation of ischemic brain injury. Although the exact mechanism is yet unknown, water diffusivity may decrease by as much as 50% in acute brain infarction. In a study of patients with acute to early subacute middle cerebral artery (MCA) territory infarcts, 38 more marked decreases in diffusivity were observed in white matter than in gray matter, with conventional diffusion-weighted imaging (DWI) underestimating the magnitude and in some cases the spatial extent of the white matter abnormalities seen by DTI.

In addition to the well-known decrease in ADC that identifies acutely infarcted gray and white matter, the relationship of new stroke lesions to major white matter tracts can also be determined by DTI, which may allow more accurate prognosis of long-term recovery or disability. 39 Furthermore, decreases in anisotropy can be observed in white matter ischemic injury, 16 and disruption or distortion of white matter tracts can also be inferred in subacute stroke patients. 40 In longer-term follow-up, DTI may also allow direct monitoring of Wallerian degeneration following ischemia, which is not possible with conventional imaging. For example, in a study of 5 patients who underwent DTI at 2 to 6 months following MCA territory infarction, evaluation of the corticospinal tract away from the area of infarction demonstrated decreased FA in all patients, despite the normal appearance of this region on conventional T2-weighted imaging in all but 1 patient. 41

Diffusion tensor imaging may also complement conventional imaging in the evaluation of traumatic head injury. In 5 patients with mild traumatic brain injuries examined within 24 hours of injury, significant reductions in anisotropy in several brain regions ipsilateral to the injury were observed when compared with the contralateral side, as well as when compared with normal controls. 42 Repeat imaging was performed in 2 of these patients at 1 month, revealing a decrease in the observed anisotropy differences at follow-up.

In a study of 21 patients with cortical contusion, significantly lower FA was observed in the white matter underlying the contusion when compared with other areas of white matter, while only nonsignificant increases in T2-weighted signal were noted in the same areas. 43 In a study of 2 patients with a history of blunt head trauma occurring 11 or 18 months before evaluation, both patients demonstrated regions of increased white matter ADC and 1 patient demonstrated ipsilateral decreased internal capsule FA, correlating with their clinical motor and neuropsychiatric deficits. 44 Notably, both of these patients had normal concurrent conventional MRI studies at the time of diffusion tensor imaging. Similarly, in a case report of a patient evaluated 18 months after a traumatic brain injury, decreased areas of anisotropy in the anterior limb of the internal capsule correlated with the patient's persistent sensory deficits, while preservation of FA in the posterior limb corresponding to the pyramidal tracts correlated with the patient's excellent recovery of motor function. 45

White matter diseases

Diffusion tensor imaging has been studied extensively in white matter disease, where the possibility of identifying abnormalities of diffusivity or anisotropy in patients with normal conventional MRI studies may lead to future clinical applications in diagnosis and disease monitoring, including response to therapy. Additionally, it has been suggested that differential changes in the major and minor eigenvalues of the diffusion tensor may be helpful in differentiating myelin loss from axonal loss in white matter disease states. 46

In DTI evaluation of patients with multiple sclerosis (MS), abnormalities of diffusion can be seen in normal-appearing white matter (NAWM). 47 In a study of subjects with relapsing-remitting MS (RRMS), decreased FA and increased ADC was observed in NAWM when compared with controls, most significantly in the corpus callosum, suggesting that DTI can identify brain injury in regions that appear normal on conventional MRI. 48 Higher ADC and lower FA within NAWM were also observed in
a larger study including patients with RRMS, secondary progressive MS (SPMS), and primary progressive MS (PPMS), where a correlation of these diffusion abnormalities with white matter lesion volume was also demonstrated. 49 While no differences in ADC or FA of NAWM were found between patients with RRMS, SPMS, and PPMS, a measure of disability did correlate positively with lesion ADC and negatively with lesion FA in patients with SPMS, suggesting that DTI may have a role in monitoring disease in advanced stages. Diffusion differences between MS phenotypes have also been observed. In a DTI study of patients with SPMS and PPMS, ADC measures in NAWM, normal appearing gray matter (NAGM), and hyperintense lesions on T2-weighted conventional MRI were significantly higher in patients with SPMS than in patients with PPMS. 50

Changes in mean diffusivity of white matter over time may also provide a marker for disease progression in cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL). 51 Additionally, increased diffusivity and decreased anisotropy have been observed in the putamen and thalamus of CADASIL patients, where thalamic diffusivity correlated with white matter diffusivity and infarct load, and inversely correlated with Mini Mental Status Examination (MMSE) score, suggesting secondary degeneration of thalamocortical pathways resulting from ischemic white matter damage. 52

In patients with amyotrophic lateral sclerosis and essentially normal conventional MRI studies, bilaterally decreased anisotropy in the anterior aspect of the posterior limb of the internal capsule has been seen when compared with controls. 53 Decreased anisotropy has also been identified within foci of periventricular T2 prolongation in B12 leukoencephalopathy, as well as in the adjacent normal-appearing white matter. 54 Finally, in a DTI study of 12 patients with conventional MRI findings consistent with reversible posterior leukoencephalopathy syndrome, increased mean diffusivity and reduced anisotropy were seen in the posterior white matter. 55 In the 1 patient re-examined after symptoms resolved, diffusivity had corrected and anisotropy had largely corrected.

Epilepsy

As many as 20% of patients with refractory epilepsy may have normal conventional MR imaging studies. In a study of 30 refractory partial seizure patients with normal conventional MRI evaluation, foci of abnormally increased diffusivity were found in 8 of 30 patients, and abnormal FA in 2 of 30 patients. 56 In 10 refractory partial seizures patients with known acquired brain lesions on conventional MRI also included in this study, foci of abnormal diffusivity corresponding to the brain lesions were noted in all 10, with 9 of 10 also demonstrating corresponding abnormalities of anisotropy. 56 In the subgroup of patients with normal MRI and abnormal DTI, 6 of the 8 diffusivity foci and 1 of the 2 anisotropy foci concurred with electroclinical localization of seizure focus in the corresponding patient, suggesting that DTI may help identify targets for surgical intervention in many patients with normal MRI studies. 56 This is supported by a case report of a patient with refractory complex partial seizures and a normal conventional MRI evaluation, in whom DTI demonstrated an area of abnormally increased diffusivity in the right orbitofrontal region that correlated with EEG data. 57 A good clinical outcome was observed in this patient following surgical resection of this area, with pathology demonstrating gliosis.

Neuropsychiatric diseases

Diffusion tensor imaging offers the opportunity to explore subtle connectivity abnormalities that may underlie many neuropsychiatric diseases and which are not evident on conventional imaging. As with white matter diseases, this may lead to future clinical applications in diagnosis and disease monitoring. Examples of white matter abnormalities identified with DTI include abnormally decreased FA in the rolandic operculum of patients with persistent developmental stuttering 58 and in the left temporoparietal region of adults with developmental dyslexia and normal conventional MRI studies. 59 In the latter study, anisotropy also correlated with a measure of reading skill. Decreased anisotropy in frontal and left temporoparietal regions has also been observed in adolescents with disruptive behavior disorder, correlating with performance on cognitive tests, suggesting abnormalities in these white matter pathways may underlie this disorder. 60 In substance abuse, reduced anisotropy within frontal white matter has been seen in cocaine-dependent patients as compared with controls, consistent with the hypothesis that decision-making deficits seen in cocaine dependence may result from impaired orbitofrontal connectivity. 61

Diffusion tensor imaging data also support abnormal cortical connectivity in schizophrenia, especially involving the frontal lobes. In a study of 14 men with schizophrenia, decreased FA in the inferior frontal white matter was associated with impulsivity and increased ADC was associated with aggression in men with schizophrenia. 62 Another DTI study demonstrated disruption in the normal pattern of connectivity between temporal and frontal brain regions in schizophrenia, as evidenced by loss of the left-greater-than-right asymmetry of FA seen in normal subjects within the uncinate fasciculus, the major white matter tract connecting these regions. 63 Diffusion tensor imaging of the corpus callosum demonstrated increased mean diffusivity and decreased FA in the splenium but not the genu of the corpus callosum in 20 schizophrenic patients compared with 25 controls, suggesting that focal disruption of commisural connectivity may also be present in schizophrenia. 64 Decreased white matter anisotropy has also been seen in prefrontal white matter, 65 as well as more diffusely, 66 in patients with schizophrenia.

Diffusion imaging may have a role in predicting response to treatment in geriatric depression. In a study of elderly patients with major depression, lower FA in the region of the frontostriatal white matter tracts was associated with lower age-adjusted remission rate following treatment with citalopram. 67

Diffusion tensor imaging has also helped advance understanding of Alzheimer disease (AD). Reduced integrity of association white matter tracts (splenium of corpus callosum, superior longitudinal fasciculus, and left cingulum) as measured by decreased lattice index, has also been found in patients with probable AD, 68 with anisotropy of the splenium correlating with MMSE score. Others have observed similar results in the posterior corpus callosum, 69,70 and significantly lower FA has also been observed in the temporal subcortical white matter and anterior and posterior cingulate bundles of AD patients. 70 More widespread increases in white matter ADC have also been reported in AD patients, as well as abnormally elevated hippocampal ADC in patients with mild cognitive impairment, many of whom may have preclinical AD. 71

Finally, diffusion imaging may come to play a role in the diagnosis and monitoring of Creutzfeldt-Jakob disease (CJD). In a study of 4 patients with CJD, half of whom had normal conventional MRI studies, reduced ADC was seen in the basal ganglia bilaterally in all patients, with additional abnormalities seen in thalamus and cerebral cortex, and with the most extensive abnormalities seen in the patient with the longest duration of symptoms. 72 A typical ADC map from a patient with CJD is shown in Figure 6.

Brain tumors

Diffusion tensor imaging and tractography offer a unique opportunity in preoperative planning for brain tumor resection, as major white matter tracts can be identified in relation to tumors, and evidence of white matter tract edema, infiltration, displacement, and disruption may be observed. 73-76 An example of fiber tractography of the corticospinal tracts in a patient with a glioma is shown in Figure 5. In addition, DTI identification of white matter tracts can be incorporated into intraoperative neuronavigation systems used during tumor resection. 77

Other research suggests that measurements of ADC may help differentiate between necrotic and viable portions of tumors, with higher ADC seen in necrotic portions of head and neck tumors, 78 as well as brain tumors, 79 opening the door for serial ADC measurements as a possible way to follow response to therapy. Apparent diffusion coefficient measurements may also be used to help differentiate tumor pathology, where the ADC of lymphoma was found to be less than that of carcinomas, carcinoma ADC less than benign solid tumor ADC, and benign solid tumor less than benign cystic lesions. 80 The same authors achieved 86% accuracy, 84% sensitivity, and 91% sensitivity in predicting malignancy by applying an ADC threshold of 1.22 * 10 -3 mm 2 /sec.

Future directions

Much work remains in understanding the underlying mechanisms of diffusion anisotropy observed in biological tissues. The relative contributions of intracellular and extracellular compartments remain to be elucidated. Additionally, the monoexponential model used at low b values begins to fail for b values greatly exceeding 1500 sec/mm 2 , where data seem to support a biexponential model. Unfortunately, multicompartment models of "fast" and "slow" diffusion do not seem to correlate with known biologic compartments and the biexponential model also fails for even larger b values, suggesting that more complex models of restricted diffusion may be required to understand diffusion at ultra-high b values. 81

Improved methods of fiber tracking also need to be developed, especially techniques to deal with the crossing of dense white matter tracts, where the tensor formalism no longer holds, and partial volume effects, resulting from the relatively low resolution obtained today with DTI. Additionally, a better understanding of noise and motion contributions to DTI is also required, 14 issues which may become clearer as faster sequences and higher field imaging are developed.

Conclusion

Diffusion tensor imaging is a rich new technique for the exploration of tissue properties at the microarchitectural level, providing a new window through which to observe development, pathophysiology, and brain connectivity. Exciting opportunities in diagnosis and disease monitoring not possible with conventional MR imaging are emerging, and as DTI becomes more widely available, many more applications will undoubtedly develop.

Acknowledgments

The author would like to thank Pratik Mukherjee for his assistance in the preparation of this manuscript, and Dan Vigneron, Duan Xu, Roland Henry, Jeffrey Berman, Mitchel Berger, Michael Geschwind, and Julie Maas for providing figures.