Diffusion Tensor Imaging in Parkinson's Disease and Parkinsonian Syndrome: A Systematic Review

Diffusion tensor imaging (DTI) allows measuring fractional anisotropy and similar microstructural indices of the brain white matter. Lower than normal fractional anisotropy as well as higher than normal diffusivity is associated with loss of microstructural integrity and neurodegeneration. Previous DTI studies in Parkinson's disease (PD) have demonstrated abnormal fractional anisotropy in multiple white matter regions, particularly in the dopaminergic nuclei and dopaminergic pathways. However, DTI is not considered a diagnostic marker for the earliest Parkinson's disease since anisotropic alterations present a temporally divergent pattern during the earliest Parkinson's course. This article reviews a majority of clinically employed DTI studies in PD, and it aims to prove the utilities of DTI as a marker of diagnosing PD, correlating clinical symptomatology, tracking disease progression, and treatment effects. To address the challenge of DTI being a diagnostic marker for early PD, this article also provides a comparison of the results from a longitudinal, early stage, multicenter clinical cohort of Parkinson's research with previous publications. This review provides evidences of DTI as a promising marker for monitoring PD progression and classifying atypical PD types, and it also interprets the possible pathophysiologic processes under the complex pattern of fractional anisotropic changes in the first few years of PD. Recent technical advantages, limitations, and further research strategies of clinical DTI in PD are additionally discussed.

Diffusion tensor imaging (DTI) allows measuring fractional anisotropy and similar microstructural indices of the brain white matter. Lower than normal fractional anisotropy as well as higher than normal diffusivity is associated with loss of microstructural integrity and neurodegeneration. Previous DTI studies in Parkinson's disease (PD) have demonstrated abnormal fractional anisotropy in multiple white matter regions, particularly in the dopaminergic nuclei and dopaminergic pathways. However, DTI is not considered a diagnostic marker for the earliest Parkinson's disease since anisotropic alterations present a temporally divergent pattern during the earliest Parkinson's course. This article reviews a majority of clinically employed DTI studies in PD, and it aims to prove the utilities of DTI as a marker of diagnosing PD, correlating clinical symptomatology, tracking disease progression, and treatment effects. To address the challenge of DTI being a diagnostic marker for early PD, this article also provides a comparison of the results from a longitudinal, early stage, multicenter clinical cohort of Parkinson's research with previous publications. This review provides evidences of DTI as a promising marker for monitoring PD progression and classifying atypical PD types, and it also interprets the possible pathophysiologic processes under the complex pattern of fractional anisotropic changes in the first few years of PD. Recent technical advantages, limitations, and further research strategies of clinical DTI in PD are additionally discussed.

INTRODUCTION
Parkinson's disease (PD) is the second most common neurodegenerative disorder after Alzheimer disease. PD generally begins with motor symptoms, including bradykinesia, rigidity, resting tremor, and postural instability. PD also includes numerous non-motor symptoms (such as cognitive impairment, depression, sleep behavioral problems, and olfactory dysfunction) (1). The classical neuropathologic hallmark of PD involves the loss of dopaminergic neurons in the substantia nigra (SN), resulting in a decreased dopaminergic output through the corticobasal ganglia-thalamocortical motor circuit, which causes dysregulation of motor functions (2). Another neuropathologic characteristic of PD is the presence of Lewy bodies (3) and neurites in both neuron bodies and axons, including aggregates of the α-synuclein protein (4). When PD begins, the complex features of motor and non-motor symptoms reflect the progression of underlying pathologies, including α-synuclein immunoreactive inclusions in Lewy bodies; loss of dopaminergic, cholinergic, serotonergic, and noradrenergic projections from the brainstem to the midbrain and basal forebrain; and, finally, to the neocortex (5). Over the course of PD, patients commonly experienced comorbid neuropsychiatric disturbances, including depression, behavioral and cognitive deficits, and dementia, which need to be differentiated from other neurodegenerative causes such as dementia with Lewy bodies (DLB) and Alzheimer's disease. On the other hand, the term "parkinsonism" refers to motor syndromes that also present in PD, including bradykinesia, cogwheel rigidity, resting tremor, a slow shuffling gait, and imbalance. The atypical causes of parkinsonism that mimic idiopathic PD include multiple system atrophy (MSA), progressive supranuclear palsy (PSP), or corticobasal syndrome (CBS), as well as drug-induced parkinsonism and others. The most common motor and non-motor syndromes of PD, and other causes of parkinsonism that are involved in PD differential diagnosis, are listed in Table 1.
In PD, the loss of dopaminergic neurons and the accumulation of Lewy bodies are typically accompanied by damage of neuroglial cells and demyelination of axons with increasing microglia concentration in extracellular spaces. It is therefore plausible that the detection of extracellular microstructural abnormalities in brain regions with dopaminergic neurons and along dopaminergic pathways might be a biomarker of incipient PD.
Diffusion Tensor Imaging (DTI), one of the magnetic resonance imaging (MRI) sequences, has been employed to measure white matter microstructural integrity in neurodegenerative diseases, as well as to visualize brain fiber connections via tractography (6). Many quantitative DTI indices (as summarized in Table 2) could be derived from a clinical DTI sequence. Fractional anisotropy (FA), radial (RD), axial (AD), and mean diffusivity (MD) have been most commonly used to describe the degree of random motion of water molecules on a microscopic scale. Specifically, FA, which measures the directionality of random water motion, has been used to probe nerve fiber arrangements, axonal integrity, and the degree of axonal myelination (64). FA is clinically feasible to deduce the microstructural integrity of brain tissues, especially preferred to oriented tissues, such as white matter and fiber-tract architectures. MD measures the magnitude of water diffusion. A high MD is thought to indicate broad cellular damages including edema and necrosis. MD is used clinically to capture these microstructural alterations in both gray and white matter tissues (65). AD measures the magnitude of diffusion along the main axis, and RD measures the magnitude of transverse diffusion. In animal studies, increased RD appears to describe myelin pathology-induced myelin thinning (66), while decreased AD indicates acute axonal injury but does not correlate with chronic axonal damage (67). However, the clinical utilities of these DTI metrics have to consider several limitations: (1) All these DTI metrics lack validations with specific neuropathology in postmortem brain. (2) The interpretation of RD/AD might be difficult in voxels containing isotropic structure. Wheeler-Kingshott and Cercignani (68) reported that fictitious changes between RD and AD could occur in areas of low anisotropy, severe pathology, partial volume, and crossing-fibers. They advised that interpretations of RD/AD should be strongly discouraged under such circumstances. (3) There is no appropriate neurobiological implication for AD increases. Linking the increased AD to axonal recovery (as a reverse response of axonal injury) remains questioned, because many studies found AD increase is associated with neurodegenerative disorders such as dementia (69). Despite these limitations, the decreased FA and increased MD have been found correlated with neuronal degeneration (70) and degeneration caused by dopamine loss (71). Further, previous studies (72,73) reported that abnormal FA values are detected in PD prior to atrophy, suggesting the usefulness of DTI measures as a biomarker of PD in clinical studies.
This study aims to review a majority of clinically employed DTI studies in investigations of PD, together with a comparison of the results from a longitudinal, early stage, multicenter clinical cohort. It is sought to prove the utilities of DTI as a biomarker of diagnosing PD, correlating clinical symptomatology, and tracking disease progression and treatment effects on PD. Advantages and challenges of clinical application of DTI in PD will be discussed further.  (41), the tracts projecting to the right pre-and primary motor cortices correlated with high tremor scores (42).
-High MD of the thalamus, SLF correlated with a worse motor speed and balance (19).
-High MD of the brainstem correlated with autonomic dysfunction during REM sleep (33).
-High MD of the SN correlated with smell loss (49).

Non-hypothesis Driven, Voxel-Wise DTI Analyses in PD
The voxel-based whole brain analysis is a computational approach to identify to what extent in brain anatomy there are significant group differences in DTI indices, via a voxelby-voxel comparison throughout the entire brain. This analysis automatically registers individual DTI maps to a template, which reduces the significant differences in brain anatomy between people. Registration algorithms can be intensity-based, for example, voxel-based analyses (VBA) (74), or tract-based, for example, tract-based spatial statistics (TBSS) (75). The advantages of the whole brain analyses are fully automatic, unbiased, unsupervised, without the need of prior hypotheses. However, the disadvantages of the whole brain approaches could be sensitive to various artifacts and technical issues, which include mis-registration of brain tissues, mis-classification of tissue types, etc. All these matters may confound the statistical analysis and either decrease the sensitivity to true DTI effects or increase the chance of false positives. Multiple studies have utilized non-hypothesis driven DTI analyses ( It is therefore conceivable that the inconsistent findings of DTI abnormality across previous studies reflect the clinical heterogeneity of the study population. In this context, it will be indeed necessary to replicate these findings in a large population, detect imaging-clinical correlations under unique types of dysfunction, and track how these findings may alter during the early course of PD.

The Parkinson's Progression Markers Initiative (PPMI) Findings
Previous studies have highlighted a very complex picture of DTI as a PD-specific biomarker. For example, some studies reported that PD exhibits significantly lower FA, but some others showed PD has higher FA than HC subjects. To address if an FA increase or decrease is a characteristic feature of PD, we investigated the temporal changes of FA along the early disease course, by analyzing data from a multicenter, longitudinal PPMI (103) cohort (from now on termed "this study"). The methodological details of data selection, MRI acquisition, image processing, and analyses of this study are described in the Supplementary Material S1-S3. Overall, the PPMI data is characterized as early diagnosed cases (diagnosed within 2 years), drug-naïve (de-novo) including 22% patients with onset at young adulthood (i.e., age ≤ 50 years). Figures below depict the findings resulted from this study, as a comparison with the numerous results published previously. Figure 1 shows statistical T-maps of FA differences between PD groups within year-1 (Y0) or more than 3 years (Y3), and demographically matched HC groups. Considering the impact of age at PD onset, the figure illustrated results of group differences separately for young-onset group (YPD, age of onset ≤ 50 years) and typical-onset group (OPD, age of onset > 50 years). In comparison between YPD(Y0) and YHC, significantly higher FA was observed in patients in multiple brain regions, including the bilateral corticospinal tract (CST), internal capsule, striatum (the putamen, pallidum, and caudate nuclei), anterior and posterior and superior corona radiata, as well as in white matter areas that lie next to the motor and supplementary motor cortices. Further, a comparison between YPD(Y3) and YHC showed that higher FA in patients remained significant in the internal capsule, striatum, corona radiata, and motor and supplementary motor white matter areas, but with a lesser extent of distribution. In the comparison between YPD and YHC groups, no region was found with lower-than-normal FA. On the other hand, the comparison between OPD(Y0) and OHC showed significantly higher FA in patients only in the bilateral CST. The comparison between OPD(Y3) and OHC showed prominently lower FA in patients in the SN, midbrain, thalamus, and multifocal non-motor areas of the white matter across all major lobes, while no area was found with higher-than-normal FA. Figure 2 depicts temporal pattern of FA changes during the early course of PD (from 0 to 79 months after the clinical diagnosis was made) in the SN and CST. For comparison, FA of the HC subjects at baseline are presented. The scatter plots are separately illustrated for the YPD vs. YHC groups and OPD vs. OHC groups. In CST, FA was significantly increased from the first year through the third year for both YPD and OPD, and later decreased to a normal level after the third year of clinical PD. In the SN, FA reduced steadily for OPD throughout the PD course.
Despite a multifocal regional pattern of FA reduction, which has been consistent with those commonly reported in literature, this study presented a robust FA increase in extensive motor areas in a large sample of the earliest and young-onset PD. The FA increase in the CST and other motor areas is not surprising because it has been also reported in previous studies using PPMI data (89,97,98) or other pilot data (80,85,92). Reduced FA is often found in neurodegenerative conditions, such as aging, Alzheimer's disease, and Parkinsonism, and is believed to be attributed to demyelination, loss of axons, and interrupted connection. It is unlikely that an FA increase is interpreted as its opposite physiological meaning (e.g., axonal regeneration, remyelination, redundant white matter networks) in neurodegenerative disorders. A meta-analysis (104) revealed that previous studies usually considered the FA increase to be due to methodological confounding factors. For example, different MRI scanners, field strength, and the number of diffusion directions could result in variability of the FA measurement and make group comparisons very difficult based on a multicenter population. Another issue is that voxels composed of    fiber populations with different spatial orientations (so called "crossing-fibers") may result in an average increase in FA. There are about 90% white matter voxels containing crossingfibers (105). Crossing-fibers may affect an anisotropy analysis and might lead to difficulties in interpreting FA increases (106). In addition, a concomitant FA increase may also be due to the variability in ROI sizes, with less or more inclusion of voxels containing surrounding isotropic neurons with low FA. However, the coexistence of FA increase and FA reduction in this study is consistent with the previous meta-analysis (107), which by accounting for heterogeneities of 39 published articles revealed FA increase in the CST and caudate nuclei, and FA decrease in the SN, corpus callosum, cingulate, and temporal areas. With an increasing number of consistent findings, an elevated FA in early PD cannot merely be attributed to methodological confounding factors but could instead indicate some pathophysiological correspondences of the disease. Possible interpretations of the FA increase include compensatory responses or excitatory reactions, which have been proposed by many fMRI (108), perfusion (109), and glucose metabolic (110) studies as a PD-related anatomical pattern of motor dysfunction (PDRD). Compensatory increased structural connectivity could be parallel to hyper-functional activation in similar motor areas (111) and are particularly activated in young aged PD (112). A previous study (98), also using PPMI data, found an FA increase in the tremor dominant (TD) phenotype, but not in the postural instability and gait difficulty (PIGD) phenotype of PD, suggesting that the compensatory white matter reorganization is more specific to the TD type in early PD. Moreover, an earlier study (73) using pilot data reported a wholebrain FA increase only in the akinetic-rigid (AR) phenotype of PD. It is therefore highly encouraged to have more investigations to validate whether the pattern of FA increase is selectively associated with a specific phenotype of motor dysfunction (i.e., TD vs. PIGD or AR subtypes), in addition to several existing pilot cohorts showing conflicting results (18,19,41,113). Although there is a lack of research to date, the pattern of coexisting FA divergency in early PD is likely also explained by the pathophysiology of inhibition (reduced white matter integrity) through direct pathways and disinhibition (increased white matter integrity) through indirect pathways in the basal ganglia (114).

DTI Changes in SN and SN Subdivisions
In contrast to the voxel-based whole-brain analysis, the hypothesis-driven region-of-interest (ROI) analysis is often used to identify DTI abnormalities in patients with PD in regions that are known to be specific to the characteristic PD pathologies. SN is one of the main sites of selective loss of dopaminergic neurons in PD, because there is already a loss of 60-80% dopaminergic neurons in the SN before PD motor symptoms emerge (115,116). Table 4 lists 32 independent studies that analyzed DTI in the total SN area and SN sub-regions in order to identify substantial DTI alterations in patients with PD. These studies were included in four previous meta-analyses and reviews (104,131,135,136 Table 4) is sensitive to capture the dopamine loss in PD. Molecular and neuropathological studies (137) of PD have highlighted that progressive loss of dopaminergic neurons is primarily involved in the pars compacta (SNc), which contains rich pigment neuromelanin formed by dopaminergic neurons, and lies in the inferior and posterior part of the SN, in contrast to the pars reticulata (SNr), which lies lateral to the SNc. Therefore, focusing the microstructural alterations on SN subregions, such as the SNc, would appear to be ideal and vital in identifying dopamine loss-related abnormality. In general practice, it is challenging to delineate the SNc on DTI image because there are no well-defined borders of this substructure. An earlier study (129) reported that reduced FA in the caudal and lateral sub-regions of the SN could completely separate patients with PD from HC with 100% diagnostic accuracy. The authors proposed that FA measurements in the caudal and lateral subregions of the SN, presumably coinciding with the histological described SNc location (ROI subdivision Type B1), have the best diagnostic value for PD. However, eight subsequent studies with the same SN subdivision (subdivision Type B) reported discrepant findings: two studies (13,132) likely replicated the findings of FA reduction in the dorsal lateral sub-region of the SN in PD; five other studies (7,8,55,104,131) found no significant FA differences between PD and HC in any SN sub-region; and one (130) reported elevated FA in all the SN sub-regions in PD. Technical difficulties outlining the sub-regions may explain these discrepant findings. To combat these issues, several studies (9,36,133) have divided the SN into two or three segments from the anterior top position through the posterior bottom position (ROI subdivision Type C). All these studies reported significantly decreased FA in the caudal (or inferior) segment of SN in patients with PD as compared with HC. Among these studies, Langley 2016 (9) used a mask of caudal zone in the SN that was previously established in a brain atlas based on magnetization transfer MRI that is sensitive to neuromelanin. This mask (i.e., the inferior portion of SN) supposedly coincides with the neuromelanin accumulation site. Another study (134) localized SN subfields using DTI tractography to segment the SN into an internal and an external part (ROI subdivision Type D), claiming that each part represents the SNc (the internal SN) and SNr (the external SN), respectively. However, no significant differences in FA or MD values between patients with PD and HC were identified in any of these subdivisions. It is likely that differences in delineating and partitioning the SN sub-regions across DTI studies are largely responsible for the inconsistent findings. Further methodological improvements are needed to ultimately demonstrate that microstructural alterations in SNc are a sensitive marker for diagnosing PD.

DTI Changes in Dopaminergic Tracts
The classical pathophysiological model of PD and recent updates have been established based on experimental and clinical studies (138)(139)(140)(141), which consists of functional disabilities of two main pathways: an "indirect circuit, " which interconnects neocortex-putamen-external pallidum/subthalamic nucleus-SN, and a "direct circuit, " which interconnects neocortex-putameninternal pallidum-SN-thalamus-neocortex. Under the conditions of PD, the reduced dopamine (results from neuronal loss in the SN) inhibits the indirect pathway, resulting in slow movement, and excites the direct pathway that may lead to unwanted movement such as resting tremor. Figure 3 (left panel) illustrates an example of these basial ganglia pathways on DTI imaging. Neuroanatomically, the dopaminergic pathway, meaning the structural connection between SN and putamen, has been assumed to contain dual effects of activities onto the indirect and direct pathways. Detecting white matter damage to these neuromodulator pathways could add to the understanding of PD pathophysiology.
Diffusion tensor tractography, as an alternative DTI-based approach, offers a possibility to detect the disrupted connections of the dopaminergic pathway and may aid the early diagnosis of PD. White matter tracts that connect the SN and striatum, namely the nigrostriatal tract (NST), are considered one of the major dopaminergic pathways, and have been shown clinically applicable in viewing anatomy of the healthy human brain (142), as well as targeting deep brain stimulation (DBS) (143,144). Although diffusion tensor tractography could not reach much detailed pathophysiologic meanings due to the limited resolution and directional information, it might still indirectly reflect a functional dysregulation of the dopaminergic circuits on a neuroanatomical basis. Figure 3 (right panel) shows an example of the nigrostriatal tractography in a healthy PPMI subject. An early DTI study (120) measured DTI indices in 11 oval ROIs that were arrayed along a line between the SN and caudate/putamen complex, and reported significantly lower than normal FA values in patients with parkinsonism at all disease stages. With the advances of DTI tractography, recent studies (13,59,89,145,146) have assessed the FA and tract connectivity profiles of the NST in identifying differences of nigrostriatal connection between patients with PD and healthy subjects. Most of these studies reported either an FA decrease (along with an increase in RD) (13,59) in PD or reduced tract connectivity (59,145,146). Some studies (13,59,60) also found that reductions in FA or tract profiles are associated with the severity of motor dysfunctions in PD patients. These consistent findings provide a possible explanation of PD pathophysiological mechanism, that is, the main motor manifestations of PD are related to the diminished connectivity of this dopaminergic circuit. Although based on a small sample, Menke et al. (147) demonstrated that the combination of volumes of the SN and the NST could achieve 100% sensitivity and 80% specificity for PD classification. Taken together, these studies suggest that DTI tractography is a promising, complementary marker of PD diagnosis. On the other hand, a tractographic analysis of the NST also benefits new PD treatment strategies that target the nigrostriatal pathway, such as deep brain stimulation, dopamine graft implantation, and infusion of glial-cell-line-derived neurotrophic factor.
Other white matter circuits that may be involved in PD pathology are of research interests. Braak et al. (3) suggested that Lewy bodies are mainly confined to the medulla oblongata/pontine tegmentum and olfactory bulb/anterior olfactory nucleus in the earliest stages of PD. Using voxel-based analysis, two studies (90,148) found PD is associated with abnormal diffusivity in white matter tracts adjacent to the olfactory sulcus. Although tractography of the olfactory tract is practically difficult in this area, using ROI drawing, some studies (92,119,122) reported low FA of the olfactory tract in PD. Another interesting approach is the tractographic analysis in corticothalamic connections of PD. The reduction of dopamine in SN results in abnormal activation of connections between the thalamus and the motor cortices. This may subsequently lead to inhibition of basal ganglia output and dysfunction in the cortical-subcortical circuits. Several studies have focused on the DTI connection of the corticothalamic tracts (14,60,85). However, diverse FA changes were reported in these studies.

DTI ASSESSMENT OF CLINICAL SYMPTOMATOLOGY AND DOPAMINE TRANSPORTER
PD is characterized as a wide variety of motor and nonmotor symptoms, while the neuropathological processes underlying these heterogeneous symptoms are not fully understood. Assessing imaging-clinical correlations in PD has been considered useful for adding further insights into neuropathological underpinnings of PD symptomatology. Table 2 summarizes the findings of correlations between available DTI metrics and clinical assessments.

DTI Correlates of Motor Dysfunction
A previous article (102) provided a thorough review of the correlations between diffusivity abnormalities and PD motor symptoms, but it lacked information about FA. Many studies, including the analyses with PPMI data, have reported significant correlations between decreased FA in the SN and increased severity of the motor symptoms, which are assessed by the motor  exams (part-III) of the Unified Parkinson's Disease Rating Scale (UPDRS) (7-12, 61, 92) or H&Y scales (113,121). However, some other studies failed to observe significant correlations. One reason for the discrepant reports of previous studies is the variation of "on-medication" or "off-medication" conditions in real-time examination of the PD patients. Dopaminergic treatments have substantial effects on the clinical measurement of motor symptoms such as UPDRS, whereas they have less effects on structural imaging such as DTI. Another complication is that, unlike quantitative imaging measurement, UPDRS is a subject measurement and depends on observers' experiences (149). Nevertheless, the consistent correlation findings from many previous studies imply that nigral FA values could offer an objective assessment of severity of the PD motor dysfunction, when clinical motor measurement is not affected by real-time treatment effects. Investigations of the correlations between DTI and subtypes of motor dysfunctions remain rudimentary. A few studies revealed significant correlations between DTI changes and severities in terms of freezing of gait (17,40), risk of falling (16), severity of postural instability and gait disturbance (PIGD) symptoms (18), motor speed and balance (19), and the degree of tremor (41,42). Some studies, using DTI for classifying PD motor subtypes (73,95), have found differentiable group effects only, but no substantial clinical correlations were observed in patients. The abnormal DTI changes in specific regions are considered useful biomarkers that characterize PD symptomology or differentiate PD phenotypes, and thus may provide an enormous potential for managing treatment directions of PD. However, it should be cautioned that the diversity of the subtype definition and the substantial instability in motor scoring might hamper the actual imaging-clinical correlations and interpretations. More reliable, objective measures of motor subtypes are necessary to validate these subtype-associated DTI changes conclusively.

DTI Correlates of Non-motor Dysfunction
Cognitive impairment is one of the major non-motor syndromes in PD. In patients with non-demented PD, the relationships between white matter integrity and global cognition, functions of various cognitive domains have been investigated in multiple studies (as illustrated in Table 2). For example, global cognitive function measured by Mini-Mental State Examination (MMSE) or Montreal Cognitive Assessment (MoCA) has been found to be associated with abnormal DTI variables in the bilateral frontal, parietal, and temporal regions, long association tracts connecting these cortices, and the hippocampus (25,26,43,45,95). Executive dysfunctions are associated with abnormal DTI variables widely distributed in the parietal and frontal regions (20-23, 43-45, 51). Declined verbal and semantic fluency and visuospatial memory are associated with abnormal diffusion variables in the parietal and frontal regions (23,24,44,51). Worsened performance of language and attention is associated with abnormal diffusivities in frontal and temporal regions and the fornix (46,47). Memory impairments are correlated with the frontal and hippocampal diffusivity abnormalities (45,48,52). Overall, the characteristic white matter abnormalities associated with cognitive impairment in PD can be summarized as follows: multiple regions are involved with a heterogeneous pattern; abnormal diffusion variables are widely distributed in white matter adjacent to cortices and limbic subcortices; the diffusion abnormalities are predominantly shown as the diffusivity changes but not the FA.
DTI correlates with other non-motor manifestations in PD have been comprehensively reviewed in a previous article (102). Briefly, DTI alterations in the thalamus (27), damaged long association tracts connecting to the frontal cortices, are associated with depression (28) and sadness (29). Diffusion changes of the hippocampus are found to be related to the impaired visuospatial memory, leading to visual hallucinations (150) in the advanced stages of PD. Diffusivity changes in the brainstem, midbrain, and pons are associated with autonomic dysfunction during a rapid eye movement (REM) sleep (33), and diffusion alterations of the limbic fornix are associated with excessive daytime sleepiness (32). However, most of these studies have been limited to small sample sizes, resulting in a low statistical power, and conflictions in terms of findings and interpretations. Further replication studies are needed to elucidate these abnormalities in more details and to truly aid the unique insights into PD symptomology.

DTI Correlates of Striatal DAT-SPECT
Dopamine transporter single photon emission tomography (DAT-SPECT) provides a meaningful measurement of striatal dopaminergic deafferentation. The best quantitative assessment on DAT-SPECT is to measure the striatal dopamine transporters binding ratio (SBR) in the left and right putamen and caudate nucleus. Studies have demonstrated substantial SBR signal loss in early PD patient relative to controls (151,152). Recent machine learning techniques (such as using the support vector machine, SVM) have shown that the SBR measures achieve high accuracies (97-100%) in predicting early PD (153,154). Additionally, the SBR has been shown strongly correlated with the number of dopaminergic neurons in substantia nigra (155). Pathological data also confirmed that quantitative measures of SBR enhance the accuracy of detecting dopaminergic neuron loss in PD (156), suggesting that SBR might be a reliable diagnostic biomarker of PD. Therefore, identifying a significant relationship between abnormal DTI and diminished SBR would support the idea that DTI alteration in the dopaminergic area is a potential marker of PD pathology. Recent findings showed a strong relationship between abnormal DTI [including decreased free water (FW) (7, 55) and increased FW (157) of the SN] and decreased putaminal SBR in PD patients. Furthermore, the progressive reduction of nigral FA has been shown correlated with the rate of SBR reduction over the first year of the PPMI patients (158). Aside from a few studies (159,160) that reported controversial findings, the strong inter-modality correlations suggest that the abnormal DTI changes in the SN are, at least indirectly, associated with the loss of dopaminergic neurons in PD. Thus, this supports the usefulness of nigral DTI measures in linking neurodegeneration to the characterized dopaminergic deficiency in PD.

DTI ASSESSMENT OF PD PROGRESSION AND TREATMENT EFFECT
Although it remains a question using the DTI as a diagnostic marker of early PD, many previous studies have reported stable, region-specific, cross-sectional correlations between DTI alterations in the dopaminergic regions and motor symptom severities, suggesting the usefulness of longitudinal DTI as a reliable biomarker for monitoring PD progression. A recent longitudinal DTI study (127) has found increased rate of DTI abnormalities (i.e., FA decrease and MD increase) in the SN of PD patients over a 19.3-month follow-up. Ofori et al. (161) reported significant FW increases in the posterior SN at 1year follow-up of PD. Further, the same authors' group (162) confirmed these longitudinal FW increases using PPMI data (103 PD at baseline, the 1st, 2nd, and 4th year) in comparison with unchanged FW in the control group (49 HC at baseline, the 1st year). Another longitudinal study (158) measuring FA and diffusivity variables of the PPMI data (122 PD vs. 50 age matched HC at baseline and 1-year follow-up) reported that PD has the highest annual rate of 3.6 ± 1.4% FA reduction in the SN, followed with a moderate rate of FA reduction in the basal ganglia, in comparison with the non-significant DTI changes in HC. Several longitudinal DTI studies identified other vulnerable regions over the PD progression: a recent study (163) found PD had a greater decrease in FA and increase in MD in the rostral brainstem, compared to controls; a 2-years longitudinal DTI study (164) reported a decrease in FA in the putamen of PD patients; another cohort (94) with 18-month follow-up of PD patients with cognitive impairment revealed greater MD increases in frontal white matter than those patients with normal cognition. These studies have presented a similar topology (mainly in the SN, also involved in the midbrain, thalamus, and to some extent to the frontal white matter) of FA decrease/MD increase over the PD course. This anatomical pattern of longitudinal DTI changes is consistent with the regional spread of Lewy body and the accumulation of Lewy neurites during PD progression (5). According to the generally consistent findings, longitudinal DTI shows a promising PD progression marker and could be valuable for monitoring and evaluating treatment effects.
Outcome measures that are most commonly used for tracking the PD treatment effects are the standard UPDRS scores (165,166) and, sometimes, the dopamine transporter imaging (167). A PPMI study (168) followed 423 patients from treatment beginning to year-5 and found that dopaminergic therapy provides significant improvements in the Movement Disorder Society revised UPDRS (MDS-UPDRS) scores and the SBR calculated from DAT-SPECT. However, these functional outcome measures may appear significantly different depending on whether they are evaluated at ON or OFF medication status. Tracking the long-term treatment effects using DTI measures could be promising as DTI is used to identify chronic responses of the brain microstructure, and thus is considered to be less affected by the ON or OFF medication status at the imaging time. Figure 4 depicts individual trajectories of nigral FA values extracted from the PPMI data (134 patients and 75 healthy subjects) over a maximum of 36 months (data information is provided in Supplementary Material S1-S3). It shows that the untreated PD patients exhibit on average a steeper decline in FA than HC subjects, while the treated PD patients maintain their respective FA levels to a decline that begins in the 3rd year of medication. This finding supports the fact that early PD is generally responsive to medications. Taken together with recent PPMI findings that nigral FA is associated with disease duration and motor rating scales (97,169), we suggest that disease modifying effects of levodopa treatment can be measured by longitudinal changes of nigral FA, at least in the first few years of treatment. Further DTI studies of PD clinical trials are needed to validate whether this is indeed true.

DTI ASSESSMENT OF DIFFERENTIAL DIAGNOSIS Differentiation Between PD and Atypical PD
It is important to diagnostically differentiate between idiopathic PD and atypical parkinsonism (aPD), including the three most common sporadic neurodegenerative syndromes: multiple system atrophy (MSA), progressive supranuclear palsy (PSP), and corticobasal syndrome (CBS), because interventional management and prognosis could be different and difficult for aPDs. DTI shows a substantial value in such differentiation.
PSP and CBS are neurodegenerative disorders characterized by abnormal tau pathology in the form of globose neurofibrillary tangles, tufted astrocytes, coiled bodies, and threads, with a predominance of 4-repeat (4R) tau isoforms (179). Both diseases provoke symptoms characteristic of PD. A recent review (180) suggests that quantitative measures and visual assessments of atrophy in the superior cerebellar peduncles (SCP) are substantial for differentiating PSP from PD, suggesting microstructural analysis of this key anatomy may provide supportive information. DTI studies (174,(181)(182)(183) showed that patients with PSP present strikingly decreased FA and increased MD in the SCP in comparison to patients with idiopathic PD. Studies (174,184) examining FA of the SCP yielded 86-94% sensitivity and 88-94% specificity for differentiating PSP from PD. Although there are few related investigations in literature, a previous study reported significantly decreased FA and increased MD in the posterior callosal truncus in CBS patients in comparison to PD. Measuring MD of the truncus yielded a 78% AUC in discriminating between CBS and PD.
In summary, PD and aPD are differentiable according to their unique anatomical distributions of microstructural abnormalities detected by reduced FA and increased MD. For example, abnormal DTI in the cerebellum and MCP may be primarily involved in MSA-P; DTI of the SCP is mostly vulnerable in PSP; and abnormal DTI in the supratentorial white matter regions is primarily involved in CBS. Additional evidence that supports these regionally specific changes in PSP and CBS are the high progressive rate of DTI abnormalities (185) and the progressive atrophy (186,187) in respective key regions of PSP and CBS observed in previous longitudinal imaging studies. For idiopathic PD, the predominant DTI changes are characterized in the SN and the basal ganglia.

Differentiation Between PD and Associated Cognitive Impairment and Dementia
Cognitive impairment, one of the non-motor symptoms, is particularly problematic in PD. Mild cognitive impairment (MCI) occurs in approximately 30% of PD patients (188). Up to 80% of late-staged PD patients eventually develop an associated dementia (PDD) (189,190). In general, subtle cognitive deficits occur in early-staged PD patients, and dementia typically occurs in elderly patients or patients in advanced stages. With matched stages and ages, PD patients with MCI (PD-MCI) show a decreased FA and increased MD compared to cognitively normal PD (PD-CN) predominantly in the frontal and interhemispheric white matter (i.e., genu and body of the corpus callosum) (45,(191)(192)(193). The PDD group, when compared to the nondemented PD group, showed FA decrease and/or MD increase in the bilateral cingulate tract (194)(195)(196), hippocampus (26), prefrontal white matter, and the genu of the corpus callosum (196,197). In addition, cognitive status, measured by MMSE or MoCA, was associated with abnormal DTI changes in the corpus callosum, anterior cingulate, and several frontal white matter areas (26,195,196,198). Moreover, Shchuz's study (199) suggested that degeneration of the nucleus basalis of Meynert measured by DTI predicts the onset of cognitive impairment, whereas a PPMI study (200) could not confirm DTI as a predictor for further cognitive decline.
The neuropathological basis of dementia in PD is not yet clear. Detectable DTI alterations in the white matter connecting to the neocortex and limbic cortex, including the frontal white matter, the corpus callosum, cingulum, and hippocampus may provide meaningful information in identifying dementia from PD. Consistent with Braak's stages V-VI (3), when α-synuclein spreads to neocortex particularly to the frontal, temporal, and limbic cortices, the abnormal DTI changes in these neuronal fibers play an important role in manifesting dementia processing in PD. On the other hand, the primary mechanisms of cognitive impairment in PD have been manifested by atrophy and reduced glucose hypometabolism in the cerebral cortices (e.g., the temporoparietal cortices) and limbic cortices (e.g., the posterior cingulate gyrus and the hippocampus) (201). White matter burden might occur secondarily or indirectly after the gray matter pathologies such as cortical accumulations of amyloid beta and tau. Further, a study (202) reported that PD with early onset of MCI (onset at PD < 1 year) exhibited lower FA and higher diffusivities in the frontal white matter than PD with late onset of MCI (onset at PD ≥1 year). In this context, whether the cognitive decline and dementia are manifested by age or dopamine deficiency needs further investigations.
A few studies (25,203) using DTI in differentiation between PDD and Lewy body dementia (LBD) found generally similar regions of reduced FA in both disorders. A few other studies (51,125) comparing PDD with Alzheimer's disease suggested that PDD may exhibit greater white matter abnormalities than Alzheimer's disease, whereas Novellin et al. (204) suggested the controversial findings. Due to the limited number investigations and the conflicting results, whether DTI can reliably differentiate PDD from LBD or Alzheimer's disease remains unknown.

DEVELOPMENTS IN DTI TECHNIQUES
Advanced Diffusion Imaging Technology Table 5 lists a comparison of imaging parameters in conventional DTI, PPMI, and advanced diffusion imaging techniques. To date, DTI is not broadly applied in clinically standard MRI examinations. In clinical practice, the conventional DTI sequence, especially for scanning patients, has to be performed within 10 min or so. The limited scan time results in problems such as low resolution, noise, artifacts, distortion and crossingfibers, and reduces qualities of the DTI image and its quantitative measures. Technical developments and limitations of DTI studies in clinical PD have been well-documented in recent review papers (107,205). New developments of diffusion imaging techniques, such as high angular resolution diffusion imaging (HARDI) (206) and diffusion spectrum imaging (DSI) (207), have shown improvements including increased signal sensitivity and resolution, gaining details of intravoxel directions and allowing a better differentiation of the crossing fibers or the joining fibers. These novel techniques are primarily used for mapping human brain connectome and brain fiber atlas building. Optimization of these sequences to be a clinically tolerable acquisition will be necessary for their clinical application.

Variables Beyond Traditional DTI Indices
Beyond the commonly employed DTI metrics including FA and diffusivities, other quantitative indices that derived from special diffusion sequence or post-image modeling provide additional information on PD biomarker research. FW is an index generated from bi-tensor diffusion model (208) and estimates fractional volume of free water within a voxel without bias from directional dependence and cellular environmental restrictions. FW expresses atrophy-based microstructural degenerations and is also used to correct partial volume effects on FA and MD measures. FW has been found more sensitive than FA in identifying abnormalities in the SN (55), significantly correlated with UPDRS (157), and persistently increased in 1, 2, and 4 years of PD (161,209). Diffusion kurtosis (210) is an index modeled from multiple b-values. Mean kurtosis (MK) estimates non-gaussian forms of diffusion and indicates microstructural complexity such as water leakage through cell membranes and myelin breakdown. It resolves the adverse effects of crossingfibers, which may influence the accuracy the FA assessment, and thus provide better efficiency in diagnosing early PD (53,211). More recently, there are increasing interests in the use of structural connectome measures to identify abnormal structural connectivity in PD or in PD motor-subtypes through networkbased statistical analysis (57,63,(212)(213)(214)(215)(216)(217). The commonly used indices derived from DTI, together with their biological interpretations and clinical correlations in PD, are summarized in Table 2. Notably, none of the aforementioned variables are specific to PD neuropathology. Postmortem studies are compulsory to confirm these imaging-pathological correlations.

DTI in Gray Matter
Voxel-based whole brain analyses provide fully automated processing and become popular in DTI studies in PD. However, these analyses have been predominantly focused on white matter structures, because the approaches highly depend on reliable co-registration and spatial normalization methods, while most image registration algorisms used for DTI analyses are established based on white matter skeleton (such as TBSS) or FA intensity, and accompany a use of FA mask (usually sets at absolute FA > 0.2) to constrain DTI analyses within the white matter. These techniques provide substantial improvements of white matter registration, but the accuracy of gray matter registration is sacrificed. Although the motor dysfunctions in PD could be examined through white matter in the pyramidal and extrapyramidal motor pathways, the non-motor symptoms are primarily driven by abnormalities in the cortical and limbic subcortical gray matter (218). Specifically, cortical deficits can be presented in early PD with mild cognitive impairments (219). Therefore, identifying PD-related gray matter microstructural abnormality would be crucial, because it can alert treatment strategies and potentially prevent patients from developing to dementia. Recent technical advantages in DTI processing, including cortical surface-based (220) registration, such as using FreeSurfer (https://surfer.nmr.mgh.harvard.edu/), as well as gray matter skeleton-based registration, such as using Gray matter-based Spatial Statistics (GBSS) (221,222), permit highly reliable DTI analyses of the gray matter. On the other hand, diffusivity variables are considered having a better sensitivity than FA in capturing neurodegenerative abnormalities in gray matter (223). Using a FreeSurfer registration, and intracortical MD measures of the PPMI data, two recent studies (224,225) with cross-sectional and longitudinal (within 1-year follow up) analyses have found that compared to controls, PD patients had significantly increased MD in the frontal and occipital cortices; among PD patients these cortical MD changes correlated with worsened cognitive performance; furthermore, the posterior-cortical (i.e., medial temporal and temporo-occipital) MD significantly correlated with the increases of serum neurofilament light chain. These DTI studies in gray matter benefits the understanding of neuropathologic mechanisms underlying PD, especially those accompanied with cognitive impairments.

CONCLUSION
This review outlines the clinical utilities of the low-cost, noninvasive diffusion tensor MRI for biomarker of diagnosing PD, correlating PD symptomatology, assessing PD progression, and differentiating atypical types of parkinsonism. The robustness of the findings provides compelling evidences that DTI may be a promising marker for monitoring PD progressing and classifying atypical PD types. Therefore, it provides outcome measures to clinical trials and helps clinicians find better patient management. But the utility of DTI for diagnosing early PD is still challenging. Another merit is that, together with the findings on PPMI data, this review presented a divergent pattern of temporal FA changes in the earliest stage of PD. In particular, FA increase is pronounced in the young onset PD and in the earliest years of PD. These observations help to improve understanding of the pathophysiologic basis (e.g., the compensatory mechanisms, excitations of the inhibitory circuits) during the earliest stages of PD.
Limitations of clinical utilities of DTI shall be mentioned: (1) Because alterations of all DTI variables generally explain the non-specific biological features, they only permit indirect interpretations of the pathogenesis underlying PD. More investigations of imaging-pathological correlations are needed to corroborate a direct implication. Clinical DTI scans are subjected to low resolution, crossing-fibers, noise, and distortions. These factors affect a precise delineation of the key substructures [e.g., SNc, olfactory bulbs and tracts, locus coeruleus, nucleus accumbens, and small fiber connections in the brainstem (226)] that might be specific to PD pathology. Advanced diffusion techniques, such as high-resolution, high filed MRI, improved distortioncorrections, as well as solutions of fiber crossings, will lead to better qualities of DTI analyses, but it is rather important to develop clinically feasible parameters upon these novel techniques. (3) DTI variables are highly dependent on scan parameters, such as MRI field strength, number of encoding directions, and maximum b values. To finally benefit the individual definition and interventions of PD, DTI measures across different MRI centers need to be harmonized, and standardized cutoff points for these DTI measures need to be established.
Further research studies are required to address (1) whether anatomically specific DTI changes are related to the core motor symptom domains (e.g., tremor, rigidity, bradykinesia, gait or postural problems) of PD and how DTI could be able to distinguish these motor subtypes; (2) how DTI change is associated with PD-risks (e.g., prodromal syndromes, risk genes) and whether this DTI change could predict further conversions to PD; (3) whether combining DTI analyses with other imaging modalities (e.g., MRI volumetric and thickness analyses, task-based and resting-state functional MRI, DAT-SPECT and neuroimaging of dopamine or alpha-synuclein, etc.), and clinical evaluations (227) could improve precisions in diagnosing early PD; (4) whether the advanced DTI analyses help the understanding of the brainstem reticular circuits and their roles in the brainstem neurotransmitter systems and how they impact the PD non-motor features (e.g., autonomic syndromes, sleep difficult and sleep behavior problems, pain, fatigue, etc.); and (5) lastly, further validation studies need to take into account of the age of PD onset, disease duration, and whether the non-motor features (especially cognitive deficits) present at the same time as the motor features, whether the PD patients are examined with MRI under an ON or OFF medication status, as well as the history and dosage of medication.

AUTHOR CONTRIBUTIONS
YZ: conception and research design, literature review, imaging data processing and analyses, results interpretation, and drafting the first manuscript. MB: conception and research design, results interpretation, language-editing, review, and critical revision of the manuscript. All authors contributed to the article and approved the submitted version.