The MRI technique diffusion tensor imaging (DTI) is reviewed along with microstructural changes associated with prodromal Alzheimer's disease (AD) as a potential biomarker for clinical applications. The prodromal stage of AD is characterized by mild cognitive impairment (MCI), representing a transitional state between normal aging and AD. Microstructural abnormalities on DTI are promising in vivo biomarkers of gray and white matter changes associated with the progression of AD pathology. Elevated mean diffusivity and decreased fractional anisotropy are consistently found in prodromal AD, and even in cognitively normal elderly who progress to MCI. However, quality of parameter maps may be affected by artifacts of motion, susceptibility, and eddy current-induced distortions. The DTI maps are typically analyzed by region-of-interest or voxel-based analytic techniques such as tract-based spatial statistics. DTI-based index of diffusivity is complementary to macrostructural gray matter changes in the hippocampus in detecting prodromal AD. Breakdown of structural connectivity measured with DTI may impact cognitive performance during early AD. Furthermore, assessment of hippocampal connections may help in understanding the cerebral organization and remodeling associated with treatment response.
Alzheimer's disease (AD) is the most common dementia beyond the age of 65. The prodromal stage in the disease process is characterized by mild cognitive impairment (MCI), representing the transition between normal aging and AD [1-4]. More and more investigators are attempting to identify biomarkers for earlier diagnoses of AD, a critical stage for preventive interventions [5, 6]. Microstructural abnormalities on diffusion tensor imaging (DTI) are promising in vivo biomarkers of gray and white matter changes associated with the progression of AD pathology. Recent studies indicate diffusivity changes are early biomarkers in predicting AD progression independent of volume loss in prodromal AD [7-11].
Diffusion tensor imaging is a measure of random motion of water molecules. The directional dependence of proton diffusion in cerebral tissue is typically quantified as fractional anisotropy (FA), and the magnitude of diffusivity is quantified as mean diffusivity (MD). Elevated MD and decreased FA are thought to reflect progressive loss of the barriers restricting the motion of water molecules in tissue compartments associated with the neuronal loss and disruption of myelin sheaths in AD [8, 12-14].
Diffusion tensor imaging is inherently a noise-sensitive and artifact-prone technique; thus, image quality assurance and robust image analysis techniques are important. Furthermore, variety of methods have been used to extract measures related to MD and FA maps, ranging from manual delineation of regions of interest to voxel/tract-based statistics to tractography. In this article, we review DTI methodologies focusing on clinical research applications in prodromal AD.
DTI Parameters and Quality Control
Parameters Characterizing the Diffusion Tensor
Diffusion tensor is reconstructed by combining diffusion measurements along at least six noncollinear spatial directions. MD and FA are the most commonly used parametric maps for characterizing the orientation-dependent proton mobility in each voxel and correlating it with the tissue architecture. In an isotropic environment, such as in the gray matter, diffusion of water molecules follows a Gaussian distribution and exhibits the same behavior in all directions. When the random motion of water molecules are directionally restricted (e.g., due to white matter tracks), the degree of this anisotropic restriction is often expressed as FA, which ranges from 0 (isotropic diffusion) to 1 (diffusion exclusively along one direction). Gray matter is characterized by isotropic diffusion; therefore, FA values of the gray matter are typically low. Thus, MD, a measure of mean area of water diffusion per unit time, is typically used to assess the microstructural integrity of gray matter. The integrity of white matter is typically assessed with both MD and FA owing to anisotropic diffusion.
In healthy white matter, water molecules diffuse more freely along the axonal fibers but are relatively restricted perpendicular to the fibers due to tightly packed axons and myelin sheets that surround them. Thus, FA is high in healthy white matter tracks . When the integrity of axons and myelin is disrupted due to pathologic processes such as AD, alterations in the diffusion characteristics of water molecules, such as increases in MD and decreases in FA, are observed in the white matter [9, 12]. These alterations in diffusion indices are not only observed in AD, but may also be observed in a variety of neurodegenerative disorders a well as MCI and even in the normal aging brain [7, 16-19].
Mean diffusivity increases with degeneration of structural barriers that restrict the Brownian motion of water molecules such as myelin and cell membranes [8, 20, 21]. Elevated MD in the hippocampus was consistently found in MCI and AD, where earliest neurodegenerative changes are observed [8, 13, 18]. Patients with AD are further characterized by increased MD in the medial and lateral temporal, and parietal lobe association cortices. This pattern of MD changes differentiates AD from other neurodegenerative dementias such as dementia with Lewy bodies (DLB) and frontotemporal lobar degeneration (FTLD) [12, 22]. Distribution of diffusivity changes in DLB and AD (Figure 1) indicated that amygdala is the only hemispheric structure involved with increases in MD in DLB, in agreement with progression of Lewy body pathology as it spreads to the temporal, occipital, and basal frontal association cortices following the involvement of amygdala. On the other hand, elevation of MD in AD involves the hippocampus and the parahippocampal gyrus early, because these are the earliest regions to be involved with the neurofibrillary pathology and neurodegeneration. White matter tracts that are on the pathways of this pathologic progression are differentially involved in these diseases. Whereas the inferior longitudinal fasciculus is involved both in AD and DLB, cingulum and fornix are only involved in AD .
Common Artifacts on DTI
Artifacts may complicate diffusion parameters and their biologic interpretation. Physiologic noise and image artifacts as well as the analytic methods may contribute to variability in findings across studies . To differentiate artifacts from tissue signal, we summarize problems that may affect the quality of DTI parameter maps such as motion, susceptibility artifacts, and eddy current-induced distortions (i.e., artifacts related with electric currents by a changing magnetic field in the conductor) (Table 1).
Table 1. The source of artifacts in diffusion tensor imaging and suggested solutions
Source of artifact
Global head motion
Keeping the subject immobile, sedation may be necessary
Physiologic motion (e.g., cardiac pulsation)
Cardiac gating the acquisition and post hoc rejection of contaminated data points
May be severe in regions adjacent to the bone and air (e.g., sinuses).
Sensitivity encoding and reversed gradient may be useful
Eddy current-induced artifacts
A rim of high anisotropy along the phase-encoding direction
Eddy current correction algorithms
Typically, there are two fundamental sources of bulk motion artifact: general head movement and physiologic motion due to cardiac induced pulsations carried into the head as a pressure wave in CSF, primarily through the ventricles . Diffusion-weighting gradients produce incoherent phase shifts sufficient for signal decay in the presence of microscopic thermal motion, thereby DTI is inherently highly sensitive to motion. A small amount of subject motion, even if it is due to flow from small vessels and cerebral spinal fluid by cardiac pulsation, can lead to a significant amount of phase shift or signal loss, which can severely affect image quality. Global head motion can be considered as a rigid body motion, a combination of global translation and rotation, which causes image blurring due to moving all spins at the same velocity and producing a linearly changing phase within the direction of the applied gradient. Cardiac pulsations are not rigid body motions and produce a nonlinearly changing phase in the spins across an image. This effect in k-space, the 2D or 3D Fourier transform of the MR image measured, is a “smearing” of the parameter maps. One common method to reduce the effects of cardiac pulsations is cardiac gating of the acquisition .
Single-shot echo planar imaging is commonly used to reduce motion sensitivity, but usually suffers from low resolution and low SNR. In addition, the transverse relaxation time constant (T2*) signal decay during the lengthy echo train leads to severe image blurring and further limits the spatial resolution of DTI images. The parameter images are further distorted due to susceptibility artifacts and are prone to slice-specific eddy current-induced distortions. Susceptibility effects are particularly severe in the brain in regions adjacent to the bone and air interphases (e.g., sinuses). Improved acquisition and reconstruction techniques may help to reduce the susceptibility effect, such as sensitivity encoding and reversed gradient [26-29]. Some distortions (e.g., stretches and shears) induced by eddy currents in the gradient coils may exhibit a rim of high anisotropy along the phase-encoding direction in DTI parameter maps. Eddy currents can be corrected by matching the nondiffusion-weighted images using cross-correlation or mutual information-based cost functions, masking signals from the cerebral spinal fluid by an extended tensor model and peripheral measurements, as well as prospective high order eddy current mitigation [30-33].
Quality Control of Parameter Maps from DTI
Given the vast amount of acquired data, visual assessment by detailed slice-by-slice inspection of parameter maps to detect potential artifacts can be extremely time consuming. Different “orthogonal” views are inspected, to determine the interslice and intravolume variances. The most common artifacts that can be identified by visual assessment are coverage problems, focal artifacts due to cardiac pulsations such as those in the brain stem, and artifacts due to head motion. According to the severity of the artifact and the location (e.g., hemispheres vs. brain stem), a decision is made on whether or not to include or exclude the scan from further analysis.
Efficient algorithms, such as calculating the standard deviation across the different diffusion images to locate cardiac pulsation artifacts and to investigate image misalignment artifacts, were proposed [34-36]. Another algorithm estimates both eddy current-induced distortion artifacts and subject movement by finding the set of parameters that minimizes residual error when fitting data to the diffusion tensor model . This algorithm is useful for improving the artifact due to head motion. Binary least square maps have been proven to be powerful options to spot additional artifacts by flagging physically implausible signals such as FA values larger than 1 or when the diffusion tensor contains negative eigenvalues [38, 39]. The weighted least square calculation is an option provided by FSL 5.0 (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/), to down-weight the implausible intensity larger than 1 on FA maps .
Methodologies in Comparing Diffusion Parameters
Interpretation and comparison of findings from DTI parameter maps in both of cross-sectional and longitudinal studies are complicated by the use of various methods of analysis, such as region-of-interest (ROI) or voxel-wise approaches with voxel-based analytic techniques (VBA) and tract-based spatial statistics (TBSS) .
Quantitative ROI-Based Techniques
Studies investigating DTI abnormalities in AD and MCI, reported reduced FA in white matter regions as well as increased MD in the hippocampus and other medial temporoparietal regions using ROI approaches[8, 17, 20, 42]. Region-of-interest drawing is often combined with manual tracing, masking, or tractography and provides an accurate quantitative measure of FA or MD from the tract of interest but is time consuming. Furthermore, ROI placement is hypothesis-dependent on the regional distribution of white matter or gray matter alterations associated with AD pathology [43, 44]. Therefore, ROI-based analyses may be useful only when the target regions or tracts are already hypothesized to show abnormalities and can be anatomically identified.
Quantitative VBA Techniques
A growing number of studies employ VBA using statistical parametric mapping (SPM) (Wellcome Department of Cognitive Neurology, http://www.fil.ion.ucl.ac.uk/) to determine the topography of groupwise FA or MD differences of microstructural changes in the entire brain . VBA is a fully automated method of analysis and can be applied as a voxel-wise investigation of whole brain diffusivity. Many imaging studies localized microstructural changes related to development, degeneration, and disease by VBA methodology [10, 11, 46-48]. However, there are currently unresolved choices for the VBA methods such as the arbitrariness of the choice of spatial smoothing extent, which inevitably affect the group differences.
Quantitative Tract-Based Analysis Techniques
Tract-based spatial statistics is an option in FSL (FMRIB Software Library, http://www.fmrib.ox.ac.uk/fsl/) software. Results can be corrected for multiple comparisons using family-wise error correction or threshold-free cluster enhancement (TFCE), a method which avoids using an arbitrary threshold for the initial cluster formation [49, 50]. Furthermore, TBSS carefully tuned nonlinear registration following it by projection onto an alignment-invariant tract template (mean FA skeleton template). The registration step, a key part of the TBSS algorithm, is designed to align each individual's FA image to a common standard space. To date, the mean FA skeleton template can be generated by those following 4 kinds of algorithms, including standard FA template, most-representative-subject FA template, study-specific FA template, groupwise atlas FA template [51-58]. Many studies have now applied TBSS to identify white matter diffusion alterations in MCI, AD as well as in normal elderly who progressed to MCI and MCI who progressed to AD with consistent results (Table 2) [53, 59, 60].
Table 2. Anatomic locations reported to show reduced FA and increased MD in patients with AD pathology
No change: “−”; mild change: “+”; moderate change: “++”.
Other domain in cognitive included language, attention/executive function, and visual–spatial processing domains.
NC, normal cognitive; MCI, mild cognitive impairment; AD, Alzheimer's disease; NC/MCI, normal cognitive subject who progressed to MCI; MCI/AD, patients with MCI who progressed to AD; HP, hippocampus; pare HP, parahippocampal grus and tract; PC, precuneus; PCC, posterior cingulum cortex; NA, not available; ST-TBSS, standard; RS-TBSS, most-representative-subject TBSS; SS-TBSS, study-specific-template; FA, fractional anisotropy; MD, mean diffusivity; ROI, region-of-interest; VBA, voxel-based analytic; TBSS, tract-based spatial statistics.
Mode of anisotropy (MO) reduction: left hippocampal(left fimbria), fornix
MO increase: superior longitudinal fasciculus, corticospinal tract cross
MD increases: left hippocampal[9, 82], baseline hippocampal, left amygdala
Evidence on DTI as a Biomarker for Neurodegeneration in Prodromal AD
Microstructure and Hippocampus in Prodromal AD
Hippocampus is one of the first brain regions to be affected by AD pathology, and microstructural alterations within hippocampus have been quantified in vivo using DTI. Mean diffusivity, as a marker of microstructure, appears to be a more sensitive marker of hippocampal integrity than macrostructural measurements with MR volumetry. In a meta-analysis comparing controls and patients with MCI, the effect size of hippocampal MD was larger than the effect size of hippocampal volume . Furthermore, MD increase in the anterior hippocampus predicted the severity of episodic memory impairment in patients with early AD better than hippocampal volumetric indices [62, 63]. Higher baseline diffusivity in the hippocampal head was associated with a greater risk of progression to AD in MCI, which may help in identifying patients with MCI who will progress to AD as well as or better than hippocampal atrophy alone .
The symmetric alterations in hippocampal diffusivity has not been consistent across studies with some reporting bilateral MD increases, others reporting asymmetric findings with greater increases in the left hippocampus . For example, left hippocampal MD was higher at baseline in patients with MCI who progressed to AD compared with patients with MCI who remained stable . In contrast, FA is not as accurate for quantifying microstructural integrity of hippocampus in AD. Regression analyses using FA values revealed similar but less pronounced alterations compared with hippocampal MD . Group differences in both MD and FA were observed even after controlling for volumetric differences in medial temporal and retrosplenial regions, indicative of microstructural damage beyond that explained by volume loss in moderate to severe AD . Therefore, low FA and high MD values may be sensitive to early changes in brain microstructure that potentially precede macrostructural changes measured with volumetric MRI.
Microstructure and Memory Circuit in Prodromal AD
Memory impairment associated with hippocampal degeneration is the earliest symptom in patients with MCI who progress to AD. Multiple regression analyses revealed left hippocampal MD increase to be the strongest predictor of verbal episodic memory performance, explaining 25% of the delayed verbal recall test variance in patients with MCI . However, atrophy and MD increases were found not only in the hippocampus, but also in the fornix and the cingulum white matter tract connecting hippocampi with precuneus/posterior cingulate cortex which may be explained by degeneration in cingulum and fornix tracts, secondary to hippocampal damage .
Another study investigated whether microstructural white matter changes can be detected in cognitively normal individuals who progressed to amnestic mild cognitive impairment (aMCI). This study indicated reductions in white matter integrity in the memory circuit (the precuneus, parahippocampal cingulum, parahippocampal gyrus, and fornix) in cognitively normal individuals who progressed to aMCI . Because fibers from the hippocampus project via the fornix to the mamillary bodies, and via cingulum to the paralimbic cortex, the FA decreases in the fornix and cingulum likely correspond to the loss of neurons in the hippocampus along with the projecting tracts. For example, the local thickness of subiculum and CA1 hippocampus fields were associated with fornix integrity, indicating AD related to hippocampal neuronal injury and degeneration of projecting axons .
Decreased FA values appear to involve white matter fibers that connect to medial temporal areas during the prodromal stage of AD. The FA values of the precuneus and left parahippocampal gyrus white matter were found to be a predictor of progression from cognitively normal to aMCI and subsequent episodic memory decline . Similarly, diffusivity characteristics related to memory function was associated with the medial temporal lobe cortical MD and inferior longitudinal fasciculus and posterior and anterior cingulum FA in elderly with no dementia . Among major white matter tracts, only FA of the posterior cingulum tract was associated with all cognitive domain functions that were investigated (memory, language, attention/executive function, and visual–spatial processing domains), consistent with the hypothesis that the posterior cingulate cortex is the main connectivity hub for cognitive brain networks  (Figure 2). The cingulum fibers have been involved in various tasks of memory such as encoding (in the hippocampus), retrieval, and recognition (in the posterior cingulate, the retrosplenial cortex, posterior and medial parietal cortex). If the integrity of neuroanatomic connectivity between brain regions of memory circuit is an important determinant of neuronal network activity, damage to neuronal connections within this circuit should have a significant impact on the effective connectivity within the network. Thus, microstructural white matter changes centered around hippocampus, cingulum, and fornix axis as a circuit for memory function may serve as a potential imaging marker of early AD-related brain changes, possibly in the preclinical stage.
Microstructure and Neuronal Circuit Remodeling in Prodromal AD
Disruption of entorhinal–hippocampal and hippocampal–cingulum pathways is associated with early decline in memory function, which recovers after deep brain stimulation in prodromal AD [10, 67, 68]. Such neuronal circuit remodeling or neuroplasticity can be defined as an experience-dependent structural or functional change in neurons, as well changes in the shape and number of cellular structures in response to a continuous demand for a specific activity. The biomarkers of neuroplasticity in the adult brain include long-term potentiation, neurogenesis, and structural remodeling of various cellular components [69-72]. There are at least two motivations for the application of DTI in axonal reorganization after treatment or rehabilitation: (1) the sensitivity of FA and MD to myelin and axonal density indicate DTI's potential utility to detect axonal reorganization [73, 74]; (2) FA and MD measurements provide a sensitive longitudinal surrogate for white matter and gray matter changes in the developing and adult brain [75-77]. Increases in FA and decreases in MD were found in the selective neuroplasticity induced by memory tasks in adults, and this process is age-dependent in the cingulate cortex, corpus callosum, and dentate gyrus . Neuroplasticity during learning can improve memory function and related microstructure changes in the inferior and superior longitudinal fasciculus, the posterior part of the cingulum bundle, and the corpus callosum . Therefore, serial DTI may be useful for a more detailed understanding of neuroplasticity processes associated with microstructural changes in AD [46, 74].
Structural connectivity assessed with DTI plays an important role in cognitive performance, and the breakdown of connectivity is an important component of early AD pathophysiology. There is increasing evidence that the differential assessment of hippocampus and tracts connecting to the medial temporal lobe are sensitive to early AD-related pathology, and their integrity is associated with cognitive decline in normal aging and risk of developing AD. DTI-based indices of diffusivity as a biomarker of microstructural integrity are independent and complementary to conventional volumetric measurements of the hippocampus as a predictor of AD. To serve as a reliable and accurate surrogate of prodromal AD, diffusion parameter maps should be quality controlled before processed and analyzed with ROI, VBA, or TBSS methods. Furthermore, longitudinal DTI studies on structural connectivity of hippocampus via the fornix and cingulum tracts are required for determining whether microstructural changes are related to declining episodic memory performance in during progression of MCI and preclinical AD. Finally, longitudinal assessments of the hippocampal structural connectivity networks may help us understand the cerebral organization and remodeling after potential treatments.