Introduction
Brain MRI is commonly performed in the diagnostic workup of parkinsonism. The main purpose of this study is to assess cerebrovascular damage for the diagnosis of vascular parkinsonism and to exclude other possible but more rare causes of parkinsonism (e.g., multiple sclerosis). It can also show abnormalities which are suggestive of neurodegenerative atypical parkinsonism (AP) [
1‐
3]. Examples include atrophy and T2 hypo-intensity of the putamen, which can be seen in the parkinsonian form of multiple system atrophy (MSA-P), while signal intensity changes of the pons (“hot cross bun” sign) or pontocerebellar atrophy can point to the cerebellar form of MSA (MSA-C). Atrophy of the midbrain (“hummingbird” sign) or signal intensity changes in the superior cerebellar peduncles are suggestive of progressive supranuclear palsy (PSP). Asymmetrical cortical atrophy is the hallmark of Corticobasal degeneration (CBD). Conventional brain MRI is usually normal or will show age-related changes in early stage Parkinson’s disease (PD), which is the most frequent cause of parkinsonism [
4]. Later on, cortical atrophy of the frontal or temporal lobe can be seen in PD.
Although certainty about the diagnosis increases during clinical follow-up, the aim of ancillary investigations is to increase certainty about the diagnosis in early disease stages, which is important for adequate patient counseling and to some extent also treatment [
5]. It has been shown that the added value of conventional brain MRI in the diagnostic workup of parkinsonism is highest in case there is uncertainty about the diagnosis [
6].
In recent years, new MRI techniques have become available for clinical practice, including diffusion-weighted imaging (DWI) and diffusion tensor imaging (DTI). Diffusion MRI quantifies the random movement of water molecules and seems to represent a quantitative measure of microstructural changes in neurodegenerative pathology, even when no abnormalities are seen on conventional MRI sequences. While fractional anisotropy (FA) estimates the degree of anisotropy, i.e., restriction of the random motion of water molecules by the normal architecture of glial tissue and fiber tracts, mean diffusivity (MD) is an averaged measure of diffusivity. Loss of microstructural integrity of brain tissue is commonly reflected by a decrease in FA and an increase in MD. Two main quantitative analyses for diffusion MRI include the region of interest (ROI) method and the automated voxel-based methods. Tract-based spatial statistics (TBSS) is an example of the automated voxel-based method. These two approaches yield complementary results, but each method has its drawbacks and does not completely reflect ongoing changes [
7].
Different patterns of microstructural changes can be identified by DTI in PD and the different forms of AP, which seem to correlate with known histopathologic changes in these diseases [
8‐
10]. Examples include increase in MD and decrease in FA of the putamen or pontocerebellar structures in MSA, and diffusional changes of the midbrain and superior cerebellar peduncles in PSP [
10]. Previous studies indicate that DTI measures of the basal ganglia, brainstem, and cerebellum can accurately identify subjects diagnosed with PD and different forms of AP [
8,
11]. Despite of the positive study results, actual application of quantitative DTI in clinical practice is limited because validated diagnostic criteria are generally lacking and no clear guidelines are available how to interpret quantitative diffusional data of the individual patient. Also, few studies evaluated brain MRI and DTI in early disease stages where the added value of brain MRI is most clinically relevant [
12,
13].
Our study objective was to evaluate whether ROI measures of DTI improve the diagnostic accuracy of conventional 3 T brain MRI in the diagnostic workup of early stage parkinsonism, to differentiate between Parkinson’s disease and neurodegenerative atypical parkinsonism.
Discussion
We evaluated the diagnostic accuracy of 3 T brain MRI and DTI to differentiate AP from PD in early stage parkinsonism. Unlike previous studies, we evaluated brain MRI performed at baseline in a cohort of patients with initial uncertain clinical diagnosis. TBSS demonstrated higher MD of the centrum semiovale, external capsule, putamen, and superior cerebellum in AP in comparison to PD. FA of the centrum semiovale was significantly lower in AP. This pattern of differences in diffusivity probably represents the summation of microstructural changes of different disease entities in the AP group. In MSA-P, MD of the left putamen, left external capsule, and superior part of the cerebellar vermis proved to be statistically significant higher in comparison to PD, with lower FA in a part of the left external capsule.
Results of the TBSS and the ROI methods showed some discrepancies, such as significantly higher MD values of the SCP and midbrain in PSP demonstrated with the ROI method but not confirmed by TBSS. For other brain structures, results were in accordance such as higher MD of the putamen in MSA-P. Results of these two methods were therefore considered complementary rather than contradictory.
The diagnostic accuracy of brain MRI to identify AP as a group was not improved when combined with the ROI measurements of MD in the putamen, midbrain, and SCP, although the AUC was slightly increased when evaluating the diagnostic accuracy to identify the subgroup of MSA-P. Disease specific diagnostic measures of DTI probably give a better estimation of the diagnostic accuracy rather grouping all the different forms of AP together. This is illustrated by our ROI analyses where differences in MD of the putamen or midbrain can be identified in MSA-P or PSP, while the averaged MD for AP as a group does not statistically differ from PD.
We found significantly increased putaminal MD values for MSA-P in comparison to PD. Diffusional changes of the putamen in MSA correspond to known underlying neuropathologic changes in the nigrostriatal system, with more severe involvement of the posterior part of the putamen compared to its anterior part in MSA-P, as has been reported in previous studies [
28‐
33]. The chosen putaminal MD cutoff threshold of 0.9 × 10
−3 mm
2/s is in accordance with other studies [
32,
33].
Increased MD values in the midbrain and SCP in PSP as compared to both MSA and PD, as we found, have also been reported previously [
34‐
38]. Histopathologic changes in PSP include damage in the cerebellar dentate nucleus and its projection fibers in the SCP [
39]. The clinical significance of damage to the SCP in PSP was found uncertain, and degeneration of the SCP appears unrelated to disease duration or typical clinical findings such as gaze palsy and postural instability [
40]. It has been reported that midbrain diffusional changes in PSP seem to correlate with disease progression [
41], while in our study, this was observed already in early disease stages.
We did not find altered diffusional measures in subcortical gray matter structures in DLB. Possibly, white matter structures in the left frontal lobe and corpus callosum could be affected in DLB, though the difference in MD and FA values between PD and DLB found by our TBSS analyses were not statistically significant. There is a debate whether DLB and dementia in PD (PDD) are the same disease entities [
42]. Subtle cognitive deficits indicating frontal lobe dysfunction are common in early stage PD, while PDD frequently occurs in late stages. Contradictory study results have been published for DTI studies comparing DLB with PDD [
43‐
46], though these discrepancies could be attributed to differences in scanning protocols and MRI field strengths. It remains to be determined whether DTI could provide diagnostic markers to identify DLB.
As many studies on DTI in parkinsonism focused on group analyses and evaluating patients in advanced disease stages [
10], the challenge now lies in clinical application of quantitative DTI in the diagnostic workup of an individual patient presenting with parkinsonism. A major advantage of TBSS over the ROI method is that it enables a hypothesis-free analysis of whole brain DTI. Although effects of misalignment due to registration and smoothing are limited, TBSS is criticized for problems with the required registration process and less reliable estimation of the diffusivity at multiple fiber orientation, such as crossing fibers. Drawbacks of TBSS are that it is only suited for the evaluation of major white matter tracts and not for gray matter structures, and diagnostic criteria for clinical use are not readily provided which hinders the evaluation of an individual patient. Quantitative measures of diffusivity of an individual patient can easily be performed with the ROI method, but for use in clinical practice validated diagnostic criteria should be applied. Error correction and standardized ROI placement is warranted for reliable and reproducible quantitative DTI analysis and this should not be cumbersome. Only few studies evaluated DTI in relation to conventional brain MRI or other advanced MRI techniques. Focke et al. studied DTI in relation to R2* and based on their results R2* seems to be of value for the diagnosis of MSA and DTI for the diagnosis of PSP [
34].
Although diffusional changes are considered to represent a quantitative measure of neurodegenerative changes in the brain, changes are notoriously difficult to interpret due to an insufficient understanding of the structural underpinnings of these changes. For example, it can be debated whether diffusional changes are the representation of the primary pathologic process, the result of a secondary consequence or age-related diffusional changes [
47]. Diffusion MRI has been used to get a better understanding of specific nonmotor signs of PD such as hyposmia and depression [
48‐
51], but those probably represent a secondary consequence. Previous study results are conflicting whether or not levadopa treatment would be of influence on changes in diffusivity of different brain structures. One study reported significant differences in putaminal ADC values between PD patients on levodopa treatment and matched untreated patients [
52], and they suggested that these differences could be attributed to the use of levodopa. However, other studies found no effect of levodopa treatment on FA or ADC values [
53,
54].
The complexity of interpreting diffusional changes is illustrated by two recently published systematic reviews, which differed in their conclusion whether DTI of the substantia nigra can be used as a diagnostic marker for PD [
8,
9]. Cochrane et al. found highly significant PD induced FA reduction in the substantia nigra [
8]. On the contrary, Schwarz et al. concluded that there is insufficient evidence for nigral DTI measures to serve as a useful diagnostic marker of PD at this point in time [
9]. Differences in studies included in these two meta-analyses as well as a variation of extracted values from included studies could explain their contradicting conclusions. In our study, FA and MD values of the midbrain in PD were comparable to MSA-P and DLB, and in our TBSS analyses, no PD-specific DTI changes could be demonstrated. It remains a debate whether or not DTI would provide a new diagnostic measure for clinical use to identify PD in the early disease stages.
There are some limitations to our study:
First, our study population was relatively small, especially in relation to the number of different MRI parameters studied. It consisted of patients presenting with predominantly hypo-kinetic symptoms and uncertain clinical diagnosis, probably explaining the majority of AP patients diagnosed with MSA-P and the low prevalence of other forms of neurodegenerative AP. It has been demonstrated that the added value of brain MRI in the diagnostic workup of parkinsonism is highest for those patients where the baseline certainty about the diagnosis is lowest [
6]. Based on initial clinical evaluation, patients with probable diagnoses of PSP, DLB, CBS, and vascular parkinsonism were excluded from the study. The classic phenotype of PSP, now called Richardson’s syndrome, is characterized by early onset postural instability and falls, supranuclear gaze palsy, and cognitive dysfunction [
55]. It is the parkinsonism form of PSP, dominated by asymmetric onset, tremor, and moderate initial therapeutic response to levodopa, which renders the differentiation with PD difficult [
56]. The same accounts for corticobasal syndrome, which is suspected when cortical dysfunction is prominent (e.g., alien limb phenomenon, cognitive decline, or behavioral abnormalities), while the differentiation with PD can be difficult in case of presence of asymmetric parkinsonism and rigidity [
57,
58]. Clinical presentation of vascular parkinsonism usually includes postural instability and falls, rather than upper limb rest tremor or bradykinesia [
59]. As a consequence of our inclusion, our study is underpowered to draw definite conclusions whether DTI is of added value for the diagnosis of separate AP subgroups. On the other side, prevalence of less frequent diseases does reflect clinical practice and illustrates the challenges for ancillary investigations to identify more rare causes of parkinsonism. The new element of our study is that we evaluated whether DTI improves the diagnostic accuracy of brain MRI to identify AP as a group in case of uncertainty about the clinical diagnosis, where it is of the most clinical relevance. Our patient cohort did not include patients diagnosed with MSA-C (clinical presentation distinct from PD and MSA-P with predominant cerebellar symptoms), which could explain that no significant diffusion differences of the MCP and pons were found in MSA-P, although diffusional changes in these structures have also been reported in MSA-P [
32,
33,
60]. Although vascular parkinsonism was beyond the scope of our study, DTI could be of value for the diagnosis of vascular parkinsonism [
61,
62]. Further studies are warranted and should include a larger sample size to evaluate the additional value of DTI for improved differentiation between the various atypical parkinsonism subtypes in the early disease stages.
Second, we did not have post mortem confirmation to reach the gold standard diagnosis and cannot fully exclude misdiagnosis. Clinical follow-up enabled us to improve certainty of the diagnosis. It has been shown that in the hands of an experienced movement disorder specialist, clinical follow-up for at least 2 years enables accurate diagnosis by evaluating the rate of disease progression, treatment response and development of any red flags [
5].
Third, interpretation of the reported quantitative MD and FA values should be done with caution because comparisons of diffusional values in different studies is difficult as scanning protocols and post processing of diffusional data lack standardization. It is known that MD and FA values vary with the MRI field strength used [
63]. As DTI is sensitive to susceptibility changes, FA and MD values are probably influenced by brain iron accumulation and calcification [
9]. Future studies need to elucidate to what extent quantitative diffusion analysis should be corrected for tissue susceptibility changes, taking the MRI field strength into account. Furthermore, fractional anisotropy is derived from the first, second, and third eigenvectors (and subsequent eigenvalues), which could provide additional measures of microstructural integrity of brain tissue. Parallel imaging and other accelerating techniques enable acquisition of high resolution DTI [
64], which could be superior for the detection of more subtle neurodegenerative changes at acceptable scanning times to enable clinical application but this needs to be determined in future prospective clinical cohort studies. In a case-control study evaluating subjects with PD, it has been reported that diffusional kurtosis imaging, which enables the quantification of non-Gaussian diffusion, is a more sensitive technique than conventional DTI for assessing tissue microstructure, even in the presence of crossing fibers [
65].
Finally, although the diagnostic accuracy of brain MRI in our study was not improved by DTI using the current analyses approach, we cannot exclude that it will prove to be of added value for the diagnostic work-up of parkinsonism while using a different methods of analysis. A previous study using a slightly different ROI approach with multiple DTI measures in the basal ganglia and cerebellum reported high accuracy in classifying patients with PD, MSA-P, PSP and control subjects [
11]. The ROI method is restricted because only a few ROIs are chosen based on a priori hypothesis and diffusional changes outside the ROI are not analyzed. Furthermore, it bears the pitfall of partial volume averaging of MD and FA measurements. Although whole brain analyses were performed with TBSS, this method is less suitable for evaluating gray matter structures as has been discussed earlier.
Machine-learning algorithms have been developed for advanced MR imaging techniques, and initial results of applying this technique to analyze DTI in a cohort of patients with parkinsonism are promising [
66]. These machine-learning techniques rely on algorithms analyzing imaging data without a priori hypotheses, based on which classifiers can be constructed for pattern recognition at the individual level [
67,
68]. Compared with a single imaging technique, the advantage of using multiple techniques is to extract more features in order to more accurately profile specific neurodegenerative pathology [
68].