Fractional Anisotropy and Mean Diffusion as Measures of Dopaminergic Function in Parkinson’s Disease: Challenging Results

Patients and Controls

The patients participated in an incidence based, long-term prospective, multidisciplinary research project on newly onset parkinsonism (the NYPUM project in Umeå, Sweden), which included DTI and SPECT imaging. The recruitment procedure and protocol of the project have previously been described [16, 17]. One hundred and forty-six patients with idiopathic PD (85 men/61 women, mean age 71±10 years) and thirty-six controls (18 men/18 women, mean age 68±7 years) were invited to perform the baseline DTI/SPECT examination.

Clinical assessment and imaging were performed on five occasions (baseline, 1-year, 3-year, 5-year, and 8-year). All invited participants did not accept to take part in the imaging study, and due to time constraints, the participants did not always perform all three examinations (DTI/ DAT SPECT/ D2R SPECT) on each occasion. The number of subjects performing DTI and DAT/D2R SPECT on each occasion is shown in Fig. 1, and the participants’ clinical features are presented in Table 1. The patients were pharmacologically untreated at baseline. At follow-ups, the antiparkinson treatment was temporarily suspended 24 h before the SPECT imaging to avoid interferences with ligand uptake.

The diagnosis was set clinically according to the UK Parkinson’s Disease Society Brain Bank criteria for idiopathic PD [18], and it was reevaluated at each follow-up. Two neurologists specialized in movement disorders had to concur on the diagnosis. The median follow-up time was 5 years, ranging from 6 months to 8 years. The controls were confirmed healthy by a clinical and MRI examination on each occasion, and none of the controls developed any signs of parkinsonism.

The study was approved by the ethical board of Umeå University and all participants gave informed consent prior to enrolment. The SPECT imaging sub-study was approved by the Swedish medical products agency (Eudra-CT no 2009-011748-20) and the local radiation safety committee at Norrlands University Hospital.

MRI scanners and gamma cameras

DTI was performed in three various MRI scanners: an “old” 1.5T scanner (1.5T_O), a “new” 1.5T scanner (1.5T_N) and a 3T scanner (3T) (all scanners manufactured by Philips Medical Systems, Best, The Netherlands). The initial SPECT was performed in a brain dedicated gamma camera, the Neurocam, which later was replaced by a general purpose hybrid gamma camera, the Infinia Hawkeye (both scanners manufactured by General Electric, Milwaukee, WI, USA). The exchange of MRI and SPECT equipment was due to routine gear updating by the hospital during the course of the study. The distribution of the combinations of the MRI scanners/gamma cameras on each imaging occasion (and for all occasions together) is presented in Table 2 together with clinical features relating to all occasions. Importantly, as diffusion measures and dopamine ligand uptake ratios may vary between equipment, data from the various MRI scanner/gamma camera combinations were analyzed separately.

Diffusion tensor imaging

The DTI in the 3T and the 1.5T_N scanner was performed using single-shot spin echo EPI sequences with the following parameters: TR = shortest, TE = 77 ms, flip angle = 90 degrees, FOV = 224*224 mm, acquisition matrix = 112*112, reconstruction matrix = 256*256, b-value = 1100 s/mm², # gradients = 16 (32 in the 1.5T_N scanner), # slices = 70 (contiguous), and thickness = 2 mm. One non-gradient (B₀) volume was also sampled. The 1.5T_O scanner used the same parameters with the following exceptions: FOV = 230*230 mm, acquisition matrix = 96*96, b-value = 1000 s/mm², # gradients = 6, # slices = 24 (contiguous), and thickness = 3 mm.

The image processing was performed using the diffusion toolbox FSL (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FDT) [19]. The images were converted to NIFTI format and the B₀-image was subjected to brain extraction using bet [20]. Each B₀-image was visually inspected to ensure that the brain was appropriately extracted. The extracted B₀-image was used in the procedure of realignment and motion and eddy current correction of the diffusion images using eddy [21], applying linear/quadratic terms for the spatial model of the eddy current induced field and linear terms for the model of the diffusion gradients. The estimation phase used a splined interpolation model, and the final resampling was performed using Jacobian modulation. Five iterations were used and data was pre-filtered with a Gaussian filter with fwhm = 1 mm. All images were visually checked after correction and no sample was disregarded based on improper correction. The DTI maps of FA and MD were generated by fitting diffusion tensors to thecorrected data using the linear least squares method of dtifit.

Regions-of-interests (ROIs) in the substantia nigra (three subareas: anterior, middle, and posterior), putamen, and the caudate (Fig. 2) were drawn on the non-diffusion weighted images using ImageJ (ImageJ 1.48, National Institutes of Health, USA, http://rsb.info.nih.gov/ij) with aid from the FA image to ensure that the ROIs did not extend into white matter. The outlining was performed by an operator specialized in DTI analysis (NL) who was blinded to the status of the subjects and to the results of the other imaging modalities. The size of the ROIs was constant for each structure. The ROIs in the striatum were outlined on the slice where the structures were most prominent and clearly defined by the external and internal capsule. In the substantia nigra, the ROIs were outlined on the slice below the level of nucleus ruber [22]. The outlined ROIs were subsequently copied to the aligned DTI maps where FA and MD were calculated for each area. FA and MD in striatum were calculated as the mean of the measures in the caudate and the putamen, accounting for the difference in area sizes.

To test the reproducibility of the ROI outlining, the ROIs were redrawn from scratch on 20% of the material by the main operator and a medical intern (WH) trained in ROI outlining of brain structures. The redrawn images were randomly selected, but in proportion to the number of total scans from each of the three scanners and the number of patients and controls. As the resolution was lower in the 1.5T_O scanner than in the 1.5T_N and 3T scanners, the reproducibility tests were performed separately for data from the 1.5T_O scanner. Thirty-three scans were included in the 1.5T_O scanner data set and forty-eight scans in the combined 1.5T_N/3T scanner data set. The two operators were blinded to the results of the original outlining and to the other operator’s outlining. Using FA from all ROIs as test parameter, the intra- and interrater correlations (Pearson) were r = 0.87 and r = 0.86 for the data set related to the 1.5T_O scanner and r = 0.89 and r = 0.87 for the data set derived from the 1.5T_N/3T scanners (P < 10^-6 for all correlations). The results confirm the reproducibility of the ROI outlining.

SPECT

The DAT SPECT was performed with ¹²³I FP-CIT and the D2R SPECT was performed with ¹²³I IBZM (both isotopes manufactured by GE Health, BV, Eindhoven, The Netherlands). The effective dose was estimated at 4.3 mSv for the DAT SPECT and 6.3 mSv for the D2R SPECT according to the manufacturer. All subjects received 200 mg potassium perchlorate p.o. before and after the SPECT investigations for thyroid protection. Details of the SPECT imaging and analysis have previously been described [3, 17]. Briefly: The Neurocam (Ncam) had three collimators with a fixed rotation radius of 12.2 cm, and the resolution at 10 cm from the collimator surface was 8.5 mm. The image data were were acquired in 128*128 matrices with a pixel size of 2.00 mm. Reconstruction was performed using filtered back projection without correction for attenuation or scatter. In the Infinia Hawkeye (InfH), the resolution was 9.0 mm with an approximate rotation radius of 15 cm. An identical image matrix was used, but data were acquired with a 1.5 zoom factor, resulting in a pixel size of 2.95 mm. The image reconstruction was done using iterative reconstruction with ordered-subset expectation maximisation. Scatter- and CT-based attenuation correction was applied.

Semi-automatic ROI analysis was made on a Xeleris Workstation (General Electric, Milwaukee, WI, USA) by an experienced nuclear medicine specialist (SJM). A set of template ROIs were applied to the most representative transverse slices with the striatal and occipital (reference area) uptake areas in the same plane, comprising a  12 mm thick portion of the stack. In the DAT SPECT images, separate ROIs were applied to the caudate and putamen. The uptake ratio was calculated as the quotient between the mean activity in the target area and the occipital reference area. For data from the Infinia Hawkeye, the ratios were corrected for rotation radius [23].

Data analysis

Five different aspects of the data were analyzed:

(1) Lateralization of the minimal putaminal DAT uptake and the most affected side regarding symptoms, and (2) lateralization in terms of co-localization (on the same side) of the minimal putaminal DAT uptake and the minimal/maximal FA/MD in the three nigral subareas (anterior/middle/posterior). Only baseline data for the patients were included; follow-up data were excluded due to the patients being medicated, jeopardizing the assessment of the lateralization of symptoms (Table 1). The DAT uptake was considered lateralized (right/left) if the side difference exceeded 5% of the left-right mean, otherwise it was considered even. The same definition was applied for FA and MD. The 5% buffer was applied to account for measurement errors.

The reason for the various analytical approaches relates to the variation in methods and patient material in similar studies [10–12], but also to the progress of the disease. The putamen is the first striatal structure that gets affected, as manifested by asymmetrical DAT uptake; an asymmetry that usually remains over time [17, 24]. Thus, this known asymmetry in putaminal DAT uptake (and its contralateral relation to the clinical symptoms) was the basis in the first three analyses, together with the fact that the diffusion measures in each nigral subarea show different sensitivity and specificity for the disease [10, 25]. As the disease progresses, most dopaminergic parts of the substantia nigra becomes engaged on both sides, and in the later stages, the entire striatum becomes involved as the caudate gets affected [26]. Furthermore, the progressive loss of nigral dopamine neurons on one side may cross-engage striatum on the other side. Thus, in the fourth and fifth analysis, involving the striatal uptake, we disregard the aspects of lateralization and nigral subareas, performing the correlations using bilateral averages.

Statistics

The analysis of lateralization (A) in the former section was performed using chi-square tests on the frequency distributions (right/left/even). As the DTI and SPECT data in this test were nominal (divided into three “laterality” groups based on the 5% -rule), data were not separated between the MRI scanner/gamma camera combinations. Hence, we assumed that within subjects, the ratio between the side difference and the right-left mean would not be severely affected by changing equipment.

In the correlation analyses of continuous data (B-E in the section on Data analysis), data from each of the four combinations of MRI scanner/gamma cameras (see Table 2) were tested separately.

The correlation analyses of (B) and (C) were performed using regression of mixed linear models with fixed effects (slope and intercept). The models accounted for attrition and the repetitive nature of the data across occasions by applying an autoregressive covariance structure (AR 1). For (D) and (E), the correlation analyses were performed in a similar manner, separately for patients and controls, but the intercept of the model was treated as a randomeffect.

The statistical analyses were carried out in IBM SPSS 22.0 0 (IBM Corp., Armonk, NY, USA). The threshold of significance was set to P < 0.001 to correct for multiple comparisons, and correlations with 0.001 < P < 0.05 were defined as having a trend towards significance.

Lateralisation

At baseline, the frequency distribution of the minimal putaminal DAT uptake and the lateralization of symptoms was significantly skewed in a contralateral manner (χ²(4) = 46.1, P < 0.001, N = 111). The frequency distribution of the minimal putaminal DAT uptake was not skewed in relation to minimal FA or maximal MD in any nigral subarea (χ²(4) < 4.2, P > 0.37, N = 111), i.e. the asymmetry of the putaminal DAT uptake did not cohere with the asymmetry of FA or MD in the substantia nigra.

Minimal DAT uptake and nigral FA and MD

In the combination of 1.5T_O/Ncam, 36 cases fulfilled the criteria of having a (lateralized) minimal putaminal DAT uptake; in the 1.5T_O/InfH combination, 74 cases; in the 1.5T_N/InfH combination, 38 cases; and in the combination of 3T/InfH, 113 cases fulfilled the criteria. There was no significant correlation between the minimal putaminal DAT uptake and FA or MD in any nigral subarea for any MRI scanner/gamma camera combination (P > 0.2).

In the augmented analysis, 19 cases demonstrated a contralateral relation between the minimal putaminal DAT and the most affected side at baseline in the combination of 1.5T_O/Ncam, 15 cases in the 1.5T_O/InfH combination, and 33 cases in the combination of 3T/InfH. No significant correlations were observed, but in the last combination, the minimal putaminal DAT uptake showed a trend towards a positive correlation with FA in the middle nigra (β ₁  = 58×10^-3, P = 0.049, N = 33), see Fig. 3. Data from the combination of 1.5T_N/InfH were not analysed as no patients were subjected to that combination at baseline (Table 2).

Side differences

There were no significant correlations between the side differences of nigral FA or MD and the putaminal DAT uptake, see Table 3, but MD in the middle nigra showed a trend towards a negative correlation with the putaminal DAT uptake in the combinations of 1.5T_O/InfH and 3T/InfH.

Averages

Striatal DAT uptake v. nigral FA and MD: No significant correlations were found in either patients or controls (Table 4 and 5).

Striatal DAT/D2R uptake v. striatal FA and MD: In patients (Table 4), there was a significant negative correlation between the striatal MD and the striatal DAT uptake for the 1.5T_O/Ncam combination. Striatal MD also showed a trend towards a negative correlation with the striatal D2R uptake in the 1.5T_O/InfH combination and with the sum of striatal DAT and D2R uptake in the combinations of 1.5T_O/Ncam and 1.5T_O/InfH. No significant correlations between striatal DAT and/or D2R uptakes and striatal FA were observed in patients (Table 4)

In controls, no significant correlations between striatal DAT and/or D2R uptakes and striatal FA or MD were found (Table 5).

Intercepts

The intercepts of the modelling of side differences can be interpreted as the side difference of FA or MD when the putaminal DAT uptake is equal on both sides. A positive intercept indicates a higher value on the right side. Three intercepts of the twenty-four regressions were significant and thirteen showed a trend towards significance (Table 3). For FA, the intercepts were positive in all subareas, whereas for MD, they were positive in the anterior nigra, but negative in the middle and posterior nigra.

The intercepts of the modelling of averages describe FA and MD at a hypothetical value of zero DAT/D2R uptakes, and a significant random effect implies that the intercepts vary to a degree that they cannot be considered to have the same mean. In patients, thirty of thirty-two regressions had a positive, significant intercept, and in 56% of the cases for FA, and in 63% of the cases for MD, the random effect was significant (Table 4). The corresponding intercepts for controls were significant and positive in thirteen of sixteen regressions, but only one single significant random effect for MD was observed (Table 5).

Fractional anisotropy in substanta nigra

The absent correlations between nigral FA and DAT uptake are similar to other findings [10, 11]. Yet, weak positive correlations between nigral FA and putaminal DAT uptake have been demonstrated in early-stage, parkinsonian patients with lateralized symptoms [12]. This fits with the correlation between FA in the middle nigra and the minimal putaminal uptake in the augmented analysis of the data from the most modern equipment (3T/InfH), see Fig. 3. Hence, although nigral FA do not correlate with DAT uptake in general, nigral FA might, to a degree, reflect the dopaminergic degeneration before the dopaminergic cell loss becomes rampant and more severe microstructural alterations occur in the nuclei.

Substantia nigra contain large dopaminergic cell bodies with thin, unmyelinated axons [1]. In contrast to white matter, where the diffusivity is primarily determined by the orientation of the axons and FA becomes a good measure of neuronal integrity [28], nigral diffusivity is likely a function of the comprehensive microstructure rather than axonal integrity. The absent correlations between nigral FA and DAT uptake in controls (Table 5) suggest that the dopamine cells only marginally contribute to the diffusivity of the overall microstructure. Still, given the dominance of dopamine cells in pars compacta and that the majority of them are lost at onset in PD [29, 30], a connection between the loss of dopamine neurons (as measured by DAT uptake) and the disintegration of the microstructural homogeneity (as measured by FA) seems plausible in PD. A possible explanation for this not occurring is the neuroinflammatory response [30]. The response would create a new diffusional topography by filling the dopaminergic void with proliferating gliotic cells and their derivatives, masking any connection between FA and the number of remaining dopamine neurons. The small change in nigral FA per unit change of striatal DAT uptake (slopes, Table 4) supports the notion that the dopaminergic contribution to the nigral diffusivity is minimal (although the correlations are non-significant), and the same picture can be observed for the only valid result of the augmented analysis, see Fig. 3.

In the modelling of the averages, in patients, there were several significant random effects for the intercepts of nigral FA (Table 4). In controls, no random effect was observed (Table 5). This suggests that the nigral inflammation causes considerable variation in FA in patients with PD. Possibly; this explains those studies that contrary to common understanding have found nigral FA to be unaltered or increased in PD compared to controls [25, 31]. Additionally, the longitudinal diffusion (a.k.a. main or axial diffusion) and the FA, which usually both are decreased in white matter in degenerative diseases, do not show similar consistent changes in the parkinsonian nigra [12, 32]. This further indicates that the pathophysiological interpretation of alterations in nigral FA is complex and not as straightforward as in white matter. Moreover, the FA value is not unique, i.e. different changes in diffusion and diffusion directions can still result in the same FA [33]. Thus, the alterations that is imposed on the substantia nigra by the disease do not necessarily transpire as changes in FA, and quite possibly, other factors than diminishing dopamine cells, for instance gliosis, are more important for how nigral FA changes in PD.

The influence of secondary, non-dopaminergic effects on the nigral diffusion can possibly also explain the varying results attained after experimental dopamine depletion: both increased and decreased nigral FA has been reported and no consistent correlation between FA and the number of remaining dopamine cells have been found [8, 13–15]. Notably, the experimental studies assessing FA after a prolonged time (several weeks) showed increased FA in the substantia nigra; a situation more similar to the human case. Given the almost complete dopaminergic eradication in animal studies, this further supports that factors other than persisting dopamine neurons heavily influence the nigral FAin PD.

Mean diffusion in substanta nigra

The results relating to side differences (Table 3) indicate that MD in the middle nigra possibly increases as the dopamine neurons vanish. Similar correlations for the middle nigra were reported in a 6-[¹⁸F] fluorolevodopa PET study [10]. However, we found no correlations between DAT uptake and nigral MD in the analysis using averages or the analysis related to minimal putaminal uptake. The observation pertaining minimal putaminal uptake is similar to others [12]. Given our large sample sizes and various analytical approaches, more consistent results would have been expected across the data from the different MRI scanners/gamma cameras and also across the different analyses, particularly for the modern MRI scanners/gamma camera. Furthermore, MD changes only slightly per unit increase of DAT uptake, and there are several random effects for the intercepts in patients (Table 4), but not in controls (Table 5). These observations indicate, similarly to FA, a minor dopaminergic contribution to the nigral diffusivity and that MD changes irregularly after the microstructural transformation induced by the disease.

Mean diffusion has not been as successful as FA in discriminating patients with PD [7, 34]. In contrast to FA, MD reflects the average diffusion magnitude, and the interpretation is not linked to the direction of the longitudinal diffusion. It is a measure of extracellular volume, and should increase if cellular density lessens. As nigral MD does not change in PD [7, 34], this may support the role of the neuroinflammatory response, in which glial cells replenish the void left by the dying dopamine neurons. It also fits with experimental results: MD increases immediately after dopamine eradication and shows a correlation to the number of remaining dopamine neurons [8], but after several weeks, MD is unchanged or decreased, and the correlation to the number of dopamine neurons is lost [13]. The latter pattern is also observed for MD in knock-out animal models of PD [15]. This course of events is difficult to capture in humans as the nigral inflammatory reaction already is fairly advanced at the point of diagnosis. Notably, diffusion measures in the substantia nigra do not seem to differ between the early and late stages of the disease [25, 35], further supporting the notion that additional loss of dopaminergic neurons do not in itself considerably affect the nigral diffusivity.

Striatal diffusion

The correlations between FA and MD and dopamine activity in striatum are similar to those in the nigra: a few negative correlations for MD in patients, all found in data from the 1.5T_O scanner, but none for FA (Table 4 and 5). The change in MD in relation to DAT and D2R uptake is small, and random effects for the intercepts are common in patients (Table 4), but not in controls (Table 5). Similar to the nigra, this indicates that other striatal entities than dopaminergic play an important role for the striatal diffusion pattern; yet, correlations may be present during certain circumstances, possibly related to the level of nigrostriatal degeneration and striatal inflammation.

There is no consistent support for striatal diffusion alterations in PD or eradicative animal models [7–9, 13]. Neither is the D2R SPECT uptake significantly altered in PD [17]. Furthermore, the dopaminergic component in the striatal DTI ROIs is relatively small, especially in comparison to the substantia nigra [30], allowing other cell structures to affect the diffusional environment. This is exemplified by striatal FA increasing in neurodegenerative conditions like Multiple sclerosis and Huntington’s disease [36–38]; diseases not primarily involving dopaminergic features in the striatum. From these perspectives, the scanty correlations for striatum are not surprising. Still, when adding the DAT and D2R uptake, the dopaminergic part increases, possibly explaining the two correlations between striatal MD and the sum of DAT and D2R uptakes. However, no such correlations were observed in data from the most modern scanners (1.5T_N and 3T, see Table 4), questioning the relevance of this observation.

Variation in clinical features; an explanatory factor?

An observation relating to nigral diffusivity is that reported correlations to DAT uptake primarily occur in the middle nigra for MD [10], including those in this study, but in the rostral and caudal parts for FA [12]. Patients in the study where correlations for MD were found in the middle nigra, and in our study, were older, had longer disease duration and more severe symptoms compared to the study finding correlations for FA in the rostral and caudal areas. Hence, the variation in subareas where correlations occur might be related to differences in disease stages, as the pace and location of the attrition varies over time within the substantia nigra [1]. In addition, the diffusional architecture varies across the nigral subareas, as demonstrated by the varying sign of the intercepts for the side differences (Table 3), which further might influence where correlations arise.

As the significant correlations that we found were primarily derived from data of the 1.5T_O scanner, we tested (post-hoc) the effect of MRI scanner/gamma camera combination on four clinical features (Table 2). We found that scans from the older scanner (1.5T_O) were associated with younger age and lower UPDRS mean scores compared to the 3T scanner, and also shorter disease duration compared to the 1.5T_N scanner. The latter is not surprising as this scanner was only used for the later follow-ups. Thus, to some degree, this indicates that the correlations might depend on developmental features of the disease. On the other hand, there was no difference in the Hoehn and Yahr stages associated with the scans of the various combinations, and additionally, age and UPDRS score did not differ between data of the old and new 1.5T scanners. Also, two correlations (trending) were actually observed for the 3T scanner (FA for the augmented analysis and MD for side differences, both in the middle nigra). The comparisons regarding UPDRS III (and to some degree the Hoehn and Yahr stages) must also be treated with some caution as the means are averaged across taxations where the patients were both OFF (baseline) and ON (follow-ups) medication. Thus, although some clinical features differed between the data sets of the combinations, there is no clear evidence for this explaining the variation in when correlations between diffusion measures and dopamine uptake arise. More detailed studies must be conducted in the future to explore the purported possibility of a stage/area dependence of the correlations. Nonetheless, this dependence does not facilitate the use of these diffusion metrics to reliably assess the nigral dopamine depletion. For a true correlation to exist and be applicable, it should be present across patients independently of stage, and longitudinally within patients as the diseaseprogresses.

Limitations

In DTI, partial volume effects from neighbouring structures can affect measurements [39]. We countered the partial volume effect by using ROIs of constant size lying entirely within the structures, trying to target the dopaminergic pars compacta on the level where it was most prominent; a procedure similar to others [22]. Still, the nigral ROIs probably contain parts of the pars reticularis as well, which could influence the diffusion values. However, this is a general concern when targeting this structure [10–12, 22]. Importantly, our nigral FA values are similar to those found by others [7]. Furthermore, given the large number of vanished dopamine cells in pars compacta in PD, and the fact that the correlations for FA were just as poor in controls with similar contributions from the reticular part and where the nucleus is a fairly intact, it is unlikely that this had a major influence on the results.

Magnetic field strengths, gradients, and resolution also affect various aspects of DTI measurements. More modern scanners also have better signal-to-noise ratio due to improved coils. Subsequently, and most fundamentally, we separated data from the various equipment combinations in the analysis; yet, the number of data from each combination, see Table 2, is substantial compared to other studies [10–12], especially for the combination of 3T/InfH (N = 139 for DAT SPECT in patients). Six-gradient DTI is today uncommon (it was the only setting in the 1.5T_O scanner) but we included these data for completeness. Still, it was data from this scanner that provided most of the correlations despite the lower resolution. Critically, the resolution of the multi-gradient modern scanners was similar to other DTI studies on PD [10, 34], without improving results, suggesting that resolution is an improbable cause for the vague correlations. The reason for the different number of gradient directions in the 1.5T_N and 3T scanner was related to time constraints; when the 3T scanner was introduced, in parallel to the 1.5T_N scanner, additional non-DTI sequences were incorporated in the MRI examiniation, forcing us to shorten the DTI protocol. Still, as the field strengths differ, data from those two scanners could not have been analysed together. Nevertheless, we applied second level modelling for the diffusion gradients when correcting the data; a setting that improves the correction when using fewer gradient directions.

DTI is suggested to be executed using at least 30 gradient directions for optimal performance [40], but many comparable studies of nigral diffusivity have used gradient numbers similar to the modern scanners in this study [7, 34]. Importantly, diffusion data from the 1.5 _N scanner applying 32 gradients did not correlate to the dopamine uptake in a single case, making that issue an implausible reason for the negative findings.

This study included longitudinal data to account for within-patient variation as the disease progresses. Particularly, those patients performing DTI in the 3T scanner and the SPECT in the Infinia Hawkeye remained in the same frame of equipment for all examinations (from baseline to their last follow-up), giving additional weight to these data. Thus, statistically, we have accounted for additional variation compared to cross-sectional studies only including baseline data [10–12]. Still, the falling-off in patients over time was significant, reducing the longitudinal structure of the data. However, such loss of patients is expected in these types of long-term, prospective studies of progressive diseases in the elderly. Essentially, the attrition and time dependence of the data were accounted for in the statistical analysis, and nonetheless, viable correlations could not be identified.

Conclusion

The results add to the debate regarding the diagnostic value of nigral MD and FA [7, 34], suggesting that these diffusion measures are not optimal for assessing the dopamine related diffusivity changes that occur in the basal ganglia in PD. Still, DTI may prove a potential supplementary tool for diagnosing PD, provided that appropriate diffusion measures for complex microstructural geometries can be developed. The bi-tensor model correcting for free water diffusion serves as a promising example [41]. This would extract the true nature of nigral diffusion alterations in PD and allow DTI to be comprehensively tested as a discriminatory tool.