Diffusion Tensor Imaging and Tractography of Human Brain Development Pratik Mukherjee, MD, PhD a,*, Robert C. McKinstry, MD, PhD b,c & Conventional MR imaging of human brain maturation & Diffusion-weighted MR imaging of human brain maturation & Diffusion tensor MR imaging of human brain maturation Basic principles of diffusion tensor imaging Directionally encoded color anisotropy maps Developmental changes in mean diffusivity of gray matter and white matter Developmental changes in the diffusion anisotropy of white matter Developmental changes in the diffusion anisotropy of cerebral cortex Brain developmental changes in the diffusion tensor eigenvalues Diffusion tensor imaging of early cerebral laminar organization & Diffusion tensor tractography of human brain maturation Basic principles of three-dimensional diffusion tensor tractography Limitations of three-dimensional diffusion tensor tractography Three-dimensional diffusion tensor tractography of human brain development Beyond the tensor: high angular resolution diffusion imaging & Emerging pediatric clinical applications Congenital brain malformations Metabolic diseases: the leukodystrophies Neurodevelopmental disorders & Acknowledgments & References Conventional MR imaging of human brain maturation MR imaging has revolutionized the noninvasive sci- entific investigation of human brain maturation as well as the clinical evaluation of disorders of the developing brain in pediatric neuroradiology. The signal intensity changes on T1-weighted and T2- weighted images during brain development are thought to result from decreases in brain water content and increases in the concentration of N E U R O I M A G I N G C L I N I C S O F N O R T H A M E R I C A Neuroimag Clin N Am 16 (2006) 19���43 Ongoing support from National Institutes of Health Grants 1R01NS046432, 2R01NS037357, 1N01NS92319, and 1P30NS048056, as well as from the Neuroradiology Education and Research Foundation is acknowledged. a Neuroradiology Section, Department of Radiology, University of California at San Francisco, San Francisco, CA, USA b Neuroradiology Section, Mallinckrodt Institute of Radiology, St. Louis, MO, USA c St. Louis Children���s Hospital, Washington University Medical Center, St. Louis, MO, USA * Corresponding author. Neuroradiology Section, Department of Radiology, University of California at San Francisco, 505 Parnassus Avenue, Box 0628, San Francisco, CA 94143���0628. E-mail address: pratik@radiology.ucsf.edu (P. Mukherjee). 1052-5149/06/$ ��� see front matter �� 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.nic.2005.11.004 neuroimaging.theclinics.com 19

macromolecules, such as myelin [1,2]. In particu- lar, T1 and T2 relaxation times become shorter as myelination progresses, allowing visualization of white matter maturation in infants and children. The T1-weighted hyperintensity and T2-weighted hypointensity of white matter myelination are first seen in the brain stem and cerebellum of preterm infants as young as 25 gestational weeks. These MR imaging features of myelination extend throughout the cerebral white matter during the first 2 years of postterm life in a stereotypical pattern, with primary sensorimotor pathways, such as the pyramidal tract and somatosensory radiations, maturing much earlier than higher order association pathways, such as those of the subcortical white matter of the prefrontal and anterior temporal regions. This regional pattern of Fig. 1. Conventional MR imaging and diffusion tensor imaging of human brain development during the first decade of postnatal life. (A) Axial slices above the level of the lateral ventricles in five normal volunteers ranging in age from 1 week old to 10 years old. (B) Axial slices through the level of the basal ganglia in the same five normal volunteer children. (Courtesy of the Brain Development Cooperative Group Images from the NIH MRI Study of Normal Brain Development with permission. Available at: http://www.brain-child.org.) 20 Mukherjee & McKinstry

white matter myelination depicted noninvasively by MR imaging matches that known from histologic studies [3���5]. Qualitative developmental milestones based on the normal temporal sequence of these T1- and T2-weighted signal intensity changes have proven clinically useful in the diagnosis of abnormal brain maturation [2]. Some of these MR imaging milestones are apparent in the top two rows of Fig. 1. The images in Fig. 1 are from the National Institutes of Health (NIH) MRI Study of Normal Brain Development, a multicenter, prospective, combined cross-sectional and longitudinal investigation that is currently in progress to map brain-behavior relations of normal children (Brain Development Cooperative Group 2005). A shortcoming of almost all prior MR imaging studies of human brain maturation is that they have been based on subjects referred for clinical neuroimaging, given the practical difficulties of scan- ning unsedated infants and young children and the ethical problems associated with using sedation in Fig. 1 (continued). 21 DTI & Tractography of Human Brain Development

children for purely research purposes. The NIH MRI Study of Normal Brain Development has demon- strated that it is feasible to obtain high-quality MR imaging scans from unsedated participants through- out infancy and childhood [6,7]. Conventional MR imaging has several fundamen- tal limitations in the evaluation of brain develop- ment, however. First, quantitative measures are not available from conventional T1- and T2-weighted imaging for more objective assessment of white matter myelination. Although T1 and T2 relaxo- metry may address this shortcoming [8���14], these T1 and T2 parametric maps can be time-consuming to acquire, often have poor inherent gray-white matter contrast, and therefore have not been widely applied clinically. Second, because the T1- and T2-weighted signal intensity changes in white mat- ter are strongly dependent on the presence of mye- lin, they cannot be used to visualize white matter tracts before the onset of myelination. Third, con- ventional MR imaging techniques provide no abil- ity to depict the orientation of fibers within white matter structures or the three-dimensional (3D) course of axonal pathways. Fourth, the conven- tional MR imaging assessment of cerebral cortical maturation is largely limited to the macroscopic assessment of cortical gyration and sulcation, with no sensitivity to the underlying microscopic events that occur during cortical development. Diffusion tensor imaging (DTI) can overcome all these limi- tations of conventional MR imaging modalities, offering a richer and more detailed window into the process of human brain maturation. Diffusion-weighted MR imaging of human brain maturation Diffusion MR imaging is sensitive to the micro- scopic motion of water molecules in biologic tis- sues [15]. In scientific research studies and in clinical assessment of the human brain, diffusion- weighted imaging (DWI) is most commonly ac- quired with a single-shot spin echo echoplanar pulse sequence using Stejskal-Tanner pulsed diffu- sion gradients. The strength of the diffusion gradi- ent is largely a function of the gradient amplitude and duration and can be expressed as the diffusion- weighting factor b in units of seconds per square millimeter. In the current practice of clinical DWI, b factors are usually in the range of 600 to 1500 s/mm2. The molecular diffusion of water causes dynamic dephasing of proton spins in the measured tissue along the orientation of the ap- plied diffusion gradient, resulting in a decrease in the measured diffusion-weighted spin echo signal. Hence, DWI signal intensity is inversely related to the magnitude of water diffusion. The apparent diffusion coefficient (ADC), a quantitative measure of the magnitude of water diffusion along the direc- tion of the applied gradient, can be calculated by comparing the diffusion-weighted signal intensity at a relatively high b value (eg, 1000 s/mm2) with the signal intensity in the absence of the diffusion gradient (ie, b = 0 s/mm2). In white matter tracts with coherent fibers organized into parallel bun- dles, water diffuses more freely along the direction of the fiber bundles than orthogonal to the bun- dles, because water mobility across the fibers is hindered by structural elements of white matter, such as the myelin sheath, the axolemma, and the neurofilaments. This diffusion anisotropy of white matter tracts can be determined by measuring ADC along multiple diffusion-encoding directions. An early DWI study of human brain maturation with diffusion-sensitizing gradients applied in two orthogonal directions reported decreasing brain ADC and increasing white matter anisotropy during the first 6 months of postnatal life [16]. A later DWI study of rat pups showed that the development of diffusion anisotropy in white matter tracts precedes the onset of myelination as detected by T1- and T2-weighted MR imaging and by histology [17]. This early phase of rising white matter anisotropy is known as ������pre- myelination,������ and may at least in part be attribut- able to nonstructural factors, such as ion channel activity in the developing axolemma [18]. Morriss and colleagues [19] examined 30 children aged 1 day to 17 years using DWI acquired in three orthogo- nal directions. They detected decreasing brain ADC and increasing white matter anisotropy over the first 3 years of postterm life. These pioneering DWI studies of brain develop- ment were limited by relatively small numbers of subjects and by the small number of diffusion- encoding directions, however. In particular, the ani- sotropy measurements in these studies underestimate the true diffusion anisotropy, because most aniso- tropy information lies in the off-diagonal elements of the diffusion tensor, to which orthogonal diffusion encoding in three or fewer directions is not sensitive [20]. Moreover, computation of the full diffusion tensor, requiring at least six diffusion-encoding direc- tions, is needed to generate rotationally invariant measures of anisotropy that are not affected by changes in head position or by intersubject differ- ences in the orientation of white matter tracts. Diffusion tensor MR imaging of human brain maturation Basic principles of diffusion tensor imaging DTI methodology has been extensively reviewed elsewhere, including the underlying mathematic 22 Mukherjee & McKinstry

theory, image acquisition, postprocessing, and vi- sualization [21���23]. The reader is referred to these sources for more detail than that presented herein. The diffusion tensor is a 3 �� 3 matrix of vectors that is used in DTI to describe the 3D distribution of water motion at each spatial location (ie, voxel) in the MR image [24,25]. Because there are six inde- pendent elements of the diffusion tensor, diffusion- weighted measurements in at least six independent directions in 3D space are required to perform DTI. The spatially averaged ADC (also known as D, ��� Dav, and mean diffusivity) can be calculated as one third of the trace of the diffusion tensor, where the trace is defined as the sum of the main diagonal ele- ments of the tensor. Several different measures of diffusion anisotropy can also be derived from the tensor, including relative anisotropy (RA), frac- tional anisotropy (FA), and volume ratio (VR). Comparing the three metrics, FA is the most sensi- tive to low anisotropy values, VR is the most sensi- tive to high anisotropy values, and RA is the most linear across the entire range of anisotropy values. The anisotropy measure A�� is equal to RA divided by the square root of 2, which places A�� on an absolute scale from a minimum of 0 to a maximum of 1 [26]. The three eigenvalues of the diffusion tensor represent the magnitude of diffusion along the three principal directions in 3D space, which are mutually orthogonal. The eigenvalue with the maximum value (the ������major eigenvalue������) is the magnitude of diffusion along the orientation in which water diffuses most freely, whereas the two other eigenvalues (the ������minor eigenvalues������) repre- sent the magnitude of diffusion along the direc- tions orthogonal to this preferred orientation. The mean of the three eigenvalues is equivalent to the mean diffusivity, and the variance of the three eigenvalues is related to the diffusion anisotropy. DTI offers two important advantages over older diffusion-weighted techniques. First, quantitative measures derived from DTI, such as mean diffusiv- ity and FA, are rotationally invariant and thus are theoretically not affected by changes in head posi- tion or the orientation of white matter tracts. Sec- ond, the orientation of white matter fibers within collimated bundles can be determined from the primary eigenvector of the diffusion tensor, which is the direction corresponding to the major eigen- value. This directional information can be visu- alized with two-dimensional (2D) directionally encoded color anisotropy images [27] or with 3D fiber tractography methods [28���30]. Directionally encoded color anisotropy maps White matter tracts of the cerebral hemispheres may be classified into three distinct types: (1) asso- ciation: those that connect two different regions of the cerebral cortex within the same hemisphere (2) projection: those that connect the cerebral cortex to subcortical structures, such as the thala- mus and spinal cord and (3) commissural: those that connect cortical regions of the left hemi- sphere with those of the right hemisphere. In gen- eral, the anisotropy values of association tracts are less than those of projection tracts, which, in turn, are less than those of commissural tracts [20]. Within the association category, the aniso- tropy of short association fibers connecting adja- cent regions of cortex, also known as subcortical ������U-fibers,������ is less than those of long association fibers running in large bundles, such as the super- ior longitudinal fasciculus (SLF) and the inferior longitudinal fasciculus (ILF). Gray matter of the cerebral cortex is thought to have anisotropy of 0 in adults, to within the limits of measurement noise, at b values not greatly exceeding 1000 s/mm2 [20,25]. The three major types of white matter tracts can also be distinguished by the direction of the axons within their fiber bundles on directionally encoded color anisotropy maps. The projection of the primary eigenvector of the diffusion tensor on each of three orthogonal axes (left-right, antero- posterior, and craniocaudal) can be encoded by different colors. In the most widely accepted di- rectional encoding scheme, the left-to-right di- rection is assigned to red, the anteroposterior dimension is assigned to green, and the cranio- caudal direction is assigned to blue [27]. This works well for differentiating large association tracts, which are usually green because they con- nect anterior and posterior cortical regions within a single cerebral hemisphere, from projection pathways, which are often blue because they con- nect superior cortical areas to inferior subcortical regions and also from commissural fibers, which appear red because of their left-to-right orienta- tion across the two hemispheres. It is important to note that DTI cannot distinguish between an- terograde and retrograde axonal directions along a single orientation for example, the corticospinal tract cannot be separated from the somatosensory radiation on the basis that in the former, the axons project from the cortex down to a subcorti- cal structure, whereas in the latter, the axons pro- ject from a subcortical structure up to the cortex. Both projection pathways appear blue on direc- tionally encoded color FA maps because both have a predominantly craniocaudal orientation. They can be differentiated using 3D DTI trac- tography, however, as explained in the sec- tion on diffusion tensor tractography of human brain maturation. 23 DTI & Tractography of Human Brain Development