Magnetic resonance T1w/T2w ratio in multiple system atrophy: A combined study with voxel-based morphometry

Diagnosis of multiple system atrophy (MSA) may be improved by using multimodal imaging approaches. We investigated the use of T1-weighted/T2-weighted (T1w/T2w) images ratio combined with voxel-based morphometry to evaluate brain tissue integrity in MSA compared to Parkinson’s disease (PD) and healthy controls (HC). Twenty-six patients with MSA, 43 patients with PD and 56 HC were enrolled. Whole brain voxel-based and local regional analyses were performed to evaluate gray and white matter (GM and WM) tissue integrity and mean regional values were used for patients classication using logistic regression. Increased mean regional values of T1w/T2w in bilateral putamen were detected in MSA-P compared to PD and HC. The combined use of regional GM and T1w/T2w values in the right and left putamen showed the highest accuracy in discriminating MSA-P from PD and good accuracy in discriminating MSA from PD and HC. A good accuracy was also found in discriminating MSA from PD and HC by either combining regional GM and T1w/T2w values in the cerebellum or regional WM and T1w/T2w in the cerebellum and brainstem. The T1w/T2w image ratio alone or combined with validated MRI parameters can be further considered as a potential candidate biomarker for differential diagnosis of MSA.


Introduction
Multiple system atrophy (MSA) is an adult-onset progressive neurodegenerative disorder featured by autonomic failure, parkinsonism and cerebellar ataxia with a prevalence of about 5 cases per 100.000. Oligodendroglial cytoplasmatic inclusions consisting of misfolded a-synuclein are required for a de nite diagnosis of MSA on postmortem examination [1]. Neuronal loss, pathologic inclusions, iron accumulation and reactive astrogliosis are observed in the striatonigral and olivopontocerebellar systems [2,3]. MSA can be clinically classi ed into the parkinsonian variant (MSA-P) and the cerebellar variant (MSA-C) based on the predominant motor features which is dependent on the distribution of pathology within the basal ganglia and cerebellum [1]. However, the predominant motor features can change with time and the variability of the severity and regional distribution of pathological process accounts for a spectrum of disease.
The differential diagnosis between MSA, particularly the parkinsonian subtype, and Parkinson's disease (PD), the most frequent α-synucleinopathy, may be di cult due to the presence of common clinical features, as demonstrated by the relatively high rate of misdiagnosis at post-mortem evaluation [4].
Previous studies suggested a role of different imaging techniques in aiding the differential diagnosis between PD and MSA. Atrophy on Magnetic Resonance Imaging (MRI) of putamen, middle cerebellar peduncle, pons, or cerebellum on conventional MRI and hypometabolism on FDG-PET in the putamen, middle cerebellar peduncle, pons, and cerebellum are already included as additional features for a diagnosis of possible MSA in the current diagnostic criteria [1]. Moreover, it has been shown that suboptimal accuracy of neuroradiological diagnosis may be improved by the use of multimodal imaging approaches and advanced MRI techniques [5].
So far, the ratio of the signal intensity of the T1-weighted and T2-weighted (T1w/T2w) MRI images, has been used as semi-quantitative measure for myelin content in gray matter [6]. Compared to other quantitative MRI techniques, it has the advantage that images can be easily acquired during routine clinical examination and without complex modeling of the MR signal, with high spatial resolution and sensitivity to neurodegenerative changes [7][8][9].
In disorders with a strong demyelinating component (as e.g. in multiple sclerosis), the T1w/T2w ratio has been found to be lowered in pathologically vulnerable regions [10,11]. However, recent studies have also pointed out on a different interpretation of the T1w/T2w ratio, suggesting that this measure may also re ect axon and dendrite density or iron content [7]. Consistently, increased T1w/T2w ratio has been recently found in the substantia nigra pars compacta of PD patients compared with healthy controls [12].
Since MSA-related pathology involves oligodendroglial and neuronal loss, with astroglial and microglial activation and increased iron content, in this study, we have performed a multimodal evaluation of tissue integrity by assessing both T1w/T2w ratio images and voxel-based morphometry (VBM) on gray and white matter (GM and WM) brain compartments. A multi-parameter analysis was performed with the aims of unveiling tissue damage in MSA patients as compared to healthy controls (HC) and PD patients and for assessing a new potential candidate MRI biomarker for differential diagnosis with PD.

Subjects
Twenty-six patients with probable MSA according to current diagnostic criteria [14 with the parkinsonian variant (MSA-P) and 12 with the cerebellar variant (MSA-C)], 43 patients with idiopathic PD and 56 healthy controls (HC) participated to the current study. Motor disability was assessed with the Uni ed MSA Rating Scale, part two (UMSARS-II) and the Uni ed PD Rating Scale, part three (UPDRS-III) in MSA and PD patients, respectively. Demographic and clinical data of enrolled subjects are reported in Table 1. The study was approved by the local Ethics committee -Comitato Etico Campania Sud -and all participants signed informed consent. The study was conducted in accordance with the Declaration of Helsinki principles.

MRI data processing
For VBM analysis tissue probabilistic maps were obtained from T1w images and used to evaluate differences in terms of grey and white matter (GM and WM) atrophy. T1w native space images of each subject were segmented into GM and WM and normalized to MNI standard space using DARTEL algorithm.
[13] Then the resulting tissue (GM/WM) probabilistic maps were modulated by the Jacobian determinants of the deformations to account for local compression and expansion due to linear and non-linear transformation [14] and then smoothed with a Gaussian kernel of 6 mm FWHM. For the group analysis a group mask was created for each tissue, averaging and then binarizing with a threshold of 0.2 all the smoothed GM/WM maps of the HC subjects. Total intracranial volume (ICV) was also calculated for each subject as the sum of the three main brain tissue volumes (GM, WM and CSF).
In order to obtain semi-quantitative maps markers of myelin content, before the validated preprocessing, [6] T1w and T2w images were corrected for intensity nonuniformity with the bias correction tool implemented in the uni ed segmentation [14] and available in SPM12. Then the T2w images were linearly registered to the T1w images using the FSL tool FLIRT [15] for estimating and applying a rigid-body a ne transformation with 6 degrees of freedom and cubic spline interpolation to minimize the WM and CSF contamination of GM voxels.
[6] T1w/T2w maps were obtained using FSLMATHS to divide the T1w volumes by the corresponding aligned T2w ones. For the group analysis the DARTEL algorithm with the same group template and deformation elds as calculated for VBM analysis were used to normalize the T1w/T2w ratio maps to the standard MNI space. During the normalization procedure, T1w/T2w maps were smoothed with a Gaussian kernel of 6 mm FWHM.

Statistical analysis
The whole brain maps (GM and WM probability tissue maps and T1w/T2w ratio maps) were compared between groups in a voxel-based full factorial analysis as implemented in SPM12. Particularly a general linear model was used considering one factor of three levels for the group (MSA, PD, HC) and two (age and sex) and three (age, sex and ICV) covariate factors respectively for T1w/T2w and GM or WM analyses. Voxels were considered signi cant with p < 0.05 after family-wise error correction (FWE) for multiple comparisons as implemented in SPM12.
A post-hoc regional analysis was also performed by extracting the mean parameter values (GM and WM probability and T1w/T2w ratio) in the voxels of detected differences in the voxel-based analysis and comparing the values between groups (MSA, PD, HC) and MSA subgroups (MSA-C, MSA-P) with a two-sample t-test after correcting for age with linear regression.
Receiver operating characteristic (ROC) curve analysis was then performed to evaluate the ability of the MRI parameters to discriminate between MSA and HC, MSA and PD and MSA-P and PD. For this analysis, a generalized linear model with binomial distribution and logit link function as setting parameters was rst computed on the regional values of GM or WM and T1w/T2w. For each ROI, the ROC curve was calculated on the obtained predictive values, allowing to calculate the corresponding area under the curve (AUC), con dence interval, and p-values. The optimal cutoff point (and corresponding sensitivity and speci city) was determined using the Youden method.
Correlations between MRI parameters and clinical variables were checked with the Spearman's rank correlation coe cient; a 5% level of signi cance was used for all tests.

Results
The three groups did not differ in age and sex distribution (Table 1). Patients with MSA and PD had similar disease duration (Table 1). Disease severity, as assessed by UPDRS-III and UMSARS, is reported in Table 1. right cluster size = 1143 voxels) and in an extended cluster in the cerebellar gray matter (cluster size = 379270 voxels) (Fig. 1a). When comparing MSA vs PD patients, the former showed reduced GM volume in two clusters of bilateral putamen (left cluster size = 119 voxels, right cluster size = 749 voxels) and in an extended area of the cerebellar gray matter (cluster size = 134350 voxels) (Fig. 1b). No signi cant differences were detected when comparing PD vs HC. Post-hoc regional analyses displayed signi cant between-group differences, between MSA-P and both PD and HC, and between MSA-C and both PD and HC (Fig. 1c).
Compared to HC, MSA patients showed reduced WM volume in a cluster extending from the cerebellar WM to the brainstem (cluster size = 413700 voxels, Fig. 2a). When compared to PD patients, MSA showed a WM volume reduction in a cluster extending from the cerebellar WM to the brainstem (cluster size = 257580 voxels, Fig. 2b).
No signi cant differences were detected when comparing PD vs HC. Regional post-hoc analyses showed a signi cant difference in WM atrophy of the cerebellum/brainstem cluster between MSA-P and both PD and HC and between MSA-C and both PD and HC (Fig. 2c).
4 HC, 1 PD and 3 MSA patients were excluded from the T1w/T2w imaging analysis due to missing 3D T2w series.
Whole brain voxel-based comparison of T1w/T2w maps did not show any difference between groups. Nonetheless, regional post-hoc analysis in the aforementioned clusters of GM or WM atrophy showed signi cant differences between MSA-P and HC and between MSA-P and PD in the right and left putamen, with increased T1w/T2w values in the MSA-P subgroup (see Fig. 3b and 3c). No signi cant differences in terms of T1w/T2w values were detected in the clusters of GM atrophy in the cerebellum and WM atrophy in the cerebellum/brainstem (Fig. 3a and 3d).
Regional values in signi cantly different clusters were also used for discriminant analysis between groups by using both T1w/T2w parameters and combined GM or WM with T1w/T2w. T1w/T2w value in the left putamen discriminated MSA vs HC, and MSA vs PD, but not MSA-P vs PD. T1w/T2w value in the right putamen discriminated MSA vs HC and MSA vs PD, but not MSA-P vs PD. Combination of regional GM and T1w/T2w in the cerebellum signi cantly discriminated MSA from HC and MSA from PD, but not MSA-P from PD. The multi-parameter (multi-variate) combination of regional GM and T1w/T2w in the left putamen signi cantly discriminated MSA from HC, MSA from PD and MSA-P from PD. The multiparameter (multi-variate) combination of regional GM and T1w/T2w in the right putamen signi cantly discriminated MSA from HC, MSA from PD, and MSA-P from PD. The multi-parameter (multi-variate) combination of regional WM and T1w/T2w in the cerebellum/brainstem signi cantly discriminated MSA from HC, MSA from PD, but not MSA-P from PD. The detailed results of the ROC analysis (AUC, sensitivity, speci city and p-values) are summarized in Table 2. After correcting regional GM or WM tissue density for age and sex, residual GM volume in the cerebellum/brainstem was signi cantly correlated with UMSARS-II (ρ=-0.515, p = 0.05) and residual WM volume in the cerebellum/brainstem was signi cantly correlated with UMSARS-I (ρ=-0.560, p = 0.037) and UMSARS-II (ρ=-0.537, p = 0.047) in MSA-P patients. There were no signi cant correlations between T1w/T2w values in any region and the clinical variables.

Discussion
In this study we performed a multi-parametric evaluation of tissue integrity assessing both T1w/T2w ratio images and voxel-based morphometry on GM and WM tissue density to investigate whole brain damage in MSA patients as compared to PD patients and HC and evaluate a potential semi-quantitative biomarker for differential diagnosis with PD. Consistently, the combined use of regional GM and T1w/T2w values in the right and left putamen showed the highest accuracy in discriminating MSA-P from PD, while the combined use of the same parameters showed a good accuracy in discriminating MSA from both PD and HC. A similar good accuracy was also found in discriminating MSA from PD and HC by either combining regional GM and T1w/T2w values in the cerebellum or regional WM and T1w/T2w in the cerebellum and brainstem.
The higher degree of putaminal and cerebellar gray matter atrophy of MSA patients largely con rm previous results [17 -19] and is in keeping with the evidence of MSA being a more aggressive and widespread disorder than PD. We also found a signi cant WM volume reduction in a cluster extending from the cerebellum to the brainstem in both MSA-P and MSA-C vs PD and HC. Consistently, few previous studies have already described WM atrophy in brainstem and cerebellum in MSA-C and MSA-P patients as compared to HC. [20][21][22] In patients with PD we found no differences in terms of atrophy when compared to HC, which is also in line with previous results obtained in patients with similar disease duration. [23] Previous studies have used the T1w/T2w ratio to detect cortical changes in patients with multiple sclerosis, showing lower values in pathologically vulnerable regions. [10,11] On the other hand, higher cortical values of the T1w/T2w ratio have been found in Alzheimer's disease [24] and Huntington's disease [25] as compared to HC, and a similar nding has been reported in a subcortical structure, namely substantia nigra pars compacta, in PD patients compared to HC. [12] Decrease in neuronal density, resulting in higher myelin proportions, or iron accumulation have been suggested to account for such previous results [25] and may also account for the increased T1w/T2w ratio found in the atrophic putamen of our MSA-P patients. Indeed, putaminal diffusivity changes have been reported corresponding to prominent neuronal loss, [26] and quantitative susceptibility mapping has con rmed increased iron deposition in the putamen of MSA-P patients. [27,28] More recently, a reduced T1w/T2w ratio in the medium cerebellar peduncle has been reported in MSA-C compared to HC, [29] suggesting that this measure can also be useful as a quantitative biomarker of myelin loss.
Indeed, this measure was also found to be correlated with the ICARS score in early MSA-C but, consistently with our results, not with UMSARS part 2 scores, possibly due to the inadequacy of the latter scale that was originally developed to characterize cerebellar ataxia and parkinsonism. [29] There are, however, some important conceptual and methodological differences between this study and ours: in fact, differently from the previous study, we adhered to the original protocol described to assess T1w/T2w ratio, [6] using 3D acquisition schemes and 1 mm 3 spatial resolution in both T1w and T2w images acquisition. This protocol was speci cally developed to analyze the entire brain, i.e., without prior assumptions on the potentially affected regions. To this more general purpose, we combined the T1w/T2w analysis with a classical VBM analysis at the same (typical) resolution of morphometric studies at 3 Tesla.
We found that UMSARS motor scores in MSA-P patients were worse in patients with greater GM and WM atrophy in the cerebellum/brainstem and, similarly, UMSARS functional scores were worse in patients with greater WM atrophy in the cerebellum/brainstem. We can speculate that correlations with such clinical parameters may re ect the progressive brainstem/cerebellar involvement during the course of MSA-P. Consistently, a similar relationship between UMSARS motor scores and cerebellar atrophy in MSA-P patients has been reported in a previous VBM study, also failing to show relationships between UMSARS scores and putaminal atrophy. [30] We must acknowledge that evaluating our MSA patients also with UPDRS and an ataxia speci c scale would have been desirable to better evaluate relationships with MRI parameters.
Further validation of the T1w/T2w ratio in MSA is needed to establish if this measure alone or combined with other validated MRI parameters can be useful as a quantitative MRI biomarker for differential diagnosis and follow-up of MSA patients.

Declarations
Author contribution SP performed MRI data processing and analysis and wrote the rst draft of the paper RM, GDS, FDS and FE critically revised MRI analyses and the nal version of the paper MCR, RE, MP, PB collected clinical data e critically revised the nal version of the paper MTP collected clinical data and wrote the nal version of the paper Figure 1 Upper panel t-maps of the voxel-wise GM comparisons respectively between HC and MSA (a) and PD and MSA (b). Statistical threshold is set to p<0.05 after FWE correction for multiple comparison. Lower panel box-plots of the GM density distribution in the signi cant clusters (from the comparison HC vs MSA) for each group, and results of the post-hoc t-tests on the regional values (c). *p<0.05, **p<0.01, ***p<0.001.

Figure 2
Upper panel t-maps of the voxel-wise WM comparisons respectively between HC and MSA (a) and PD and MSA (b). Statistical threshold is set to p<0.05 after FWE correction for multiple comparison. Lower panel box-plots of the WM density distribution in the signi cant cluster (from the comparison HC vs MSA) for each group, and results of the post-hoc t-tests on the regional values. *p<0.05, **p<0.01, ***p<0.001.