A meta-analysis of diffusion tensor imaging of substantia nigra in patients with Parkinson’s disease

Parkinson’s disease (PD) is a common neurodegenerative disease characterized by severe, selective loss of pigmented neurons in the substantial nigra (SN). Previous studies have indicated that such loss could be detected by diffusion tensor imaging (DTI). Here, we try to consolidate current DTI data to both quantitatively determine the imaging changes in SN, as well as explore the potential use of DTI for PD diagnosis. Fourteen research articles are included in this meta-analysis, each obtained by searching PubMed, EMBASE, or Cochrane library database dated until July 2017. The articles contain 14 trials with 298 total PD patients and 283 healthy controls (HCs). The results show not only significantly lower FA values of SN in PD compared to that of HCs (WMD = −0.02, 95% CI = [−0.03, −0.02], p < 0.00001), but also a significantly higher MD in PD compared to HCs (WMD = 0.05, 95% CI = [0.03, 0.07], P < 0.0001). This indicates that the sharp difference detected between PD patients and HCs can be detected by DTI. By further analyzing the heterogeneity, we found that FA measurement of SN could be potentially used as a surrogate, noninvasive diagnostic marker toward PD diagnosis.

. The aim of this study was to integrate published data to estimate the diagnostic threshold of values by DTI using standard meta-analysis methods.

Study Characteristics.
Nigral MD changes. Thirteen studies (282 PD patients and 267 HCs) reported nigral MD changes that showed a significant MD increase in PD patients compared to HCs, with an estimated WMD of 0.05 (95% CI = [0.03, 0.07], p < 0.0001) (Fig. 3). A moderate-level of heterogeneity was recorded (I 2 = 58%). The random-effect model was used to calculate the pooled mean effect size.
Subgroup analysis of FA and MD changes. In terms of improved signal to noise ratio and increased image resolution, higher field strength magnets were generally considered to be more desirable and may have had an influence on imaging results. We therefore conducted a subgroup analysis based on differing field strengths. Among the thirteen studies with reported FA changes, nine studies used 3.0 Tesla. Analysis of the studies that utilized 3.0 Tesla field strength indicated a significant difference on FA changes between PD patients and HCs, with an estimated WMD −0.03 (95% CI = [−0.03, −0.02], p < 0.00001). Moreover the four studies that used a 1.5 Tesla field strength also documented a significant difference on FA changes in the meta-analysis, with an estimated   4). Among the thirteen studies that have reported MD changes, the meta-analysis of eight studies using 3.0 tesla field strengths and five studies using 1.5 tesla field strengths also indicated significance, with an estimated WMD of 0.05 (95% CI = [0.02, 0.08], p = 0.003) and 0.06 (95% CI = [0.03, 0.08], p < 0.0001), respectively (Fig. 5).
Sensitivity analyses. Sensitivity analyses were performed by removing each study in turn and re-analyzed.
No changes of significant differences were documented on DTI measurements (FA, MD) between the PD patients and HCs.

Discussion
Current meta-analysis did find significantly lower FA values and higher MD values in PD patients over HCs. By means of analyses of heterogeneity, sensitivity, and publication bias, an obvious variation in results was found in the meta-analysis for PD-induced nigral MD changes, so performed a measurement of FA in SN that could be a potential surrogate, non-invasive marker for precise diagnosis of PD. To date, no DTI technique has been used to detect preclinical or early PD diagnosis, but one study on rodent model of PD showed reduced FA in the SN at its early stage that was correlated with the loss of SN dopamine neurons 7 . This result could provide us with clues to further investigate the screening of preclinical PD via DTI measurements.
There is a possibility that imaging heterogeneity in this meta-analysis could have arisen from differences in methodology, data acquisition, field strength of magnet, etc. across the 14 chosen studies. Indeed, such heterogeneity has also been found in this current meta-analysis concerning PD diagnosis, likely attributable to the different stages or severity of disease progression in patients imaged. These factors can unavoidably affect the homogeneity of imaging results. A recent study of nigral FA changes in PD patients couldn't find any correlation between PD severity and FA changes 23 . Another possible source of heterogeneity in the data include the  conversion of median averages and accompanying graphs into mean or variance via conversion tools. While we did not find publication bias in this meta-analysis, we cannot rule out that some biases went unreported in the sourced papers.
Regarding analysis by MRI imaging, the delineation of the SN was not uniform in included studies. For example, some studies focused their analysis on a sub-region of SN, while others used the whole SN as Region of Interest (ROI) 15,18 . If SN screening was based on ROI, the averaged value from SN couldn't be compared among different studies. Prakash et al. 18 described that lower FA and higher ADC were found at the left rostral SN compared to the right side. The subjective delineation of SN on T2-weighted MRI images resulted in most of the variation in the studies 22 . The primary pathology of PD is the loss of neuromelanin-generating dopaminergic neurons 24 that can be manifested by MRI as hypodensity 25 , used by Langley et al. 13 to depict the SN for screening. However, hypodensity can also be observed by iron deposition on SN in T2-weighted images. Defining the ROI of such scans could be difficult for DTI data analysis 25 . The SN volume delineated by neuromelanin sensitive MRI and T2w images differs not only in signal characteristics but also spatially 13 . The statuses of patients either during  Two previous meta-analysis on DTI were published in 2013. The first analysis 24 focused on the diagnostic value of DTI, with four of nine studies focusing on SN and other regions focusing on the globuspallidus, anterior olfactory structures, and cerebellar peduncles. The second analysis 22 included 12 studies, eight of which have been included in this current meta-analysis, four having been excluded due to confounding components of each. Of the four excluded, three screened multiple sub-regions of SN 23,26,27 , making data collection difficult, and one focused on diffusion krutosis imaging (DKI) rather than DTI 28 . Our current study used careful review and evaluation to collect additional qualified studies that focus their analysis on the SN.
The following limitations of our study should be acknowledged: i) the limitation from the original data, we couldn't further analyze with other impact factors such as different ages, genders, with or without target medications, and severity of disease itself. ii) The standard for study inclusion varied study by study, publication biases could still exist in our current study. iii) We couldn't evaluate the sensitivity and specificity of the measure currently. Future high-quality and multicenter studies are warranted.
To date, we haven't found any studies in the literature on using the DTI technique as a direct marker for PD diagnosis, preclinical or otherwise. Our meta-analysis showed that significantly lower FA values in SN could potentially be used as diagnostic marker for PD.

Methods
Literature search. Current meta-analysis was based on the guidelines of observational studies in epidemiology (MOOSE) 29 . PubMed, EMBASE, and Cochrane library were the sources of database used in our literature search dated until July 2017. Search terms used were "Parkinson's disease", "PD", "Diffusion Tensor Imaging", "DTI", "Substantial Nigra", "SN", followed by the combinations of keywords and synonyms without language restrictions. Related references from search-out studies were also listed and reviewed to avoid possible overlook by our study. All papers have been screened by two authors who would be consensus or if the disagreement happened, the third author was consulted to decide if the study should be included or not. All included studies have been assessed by the Newcastle-Ottawa Scale (NOS) that was recommended by the Cochrane collaboration to guarantee the quality of each study 30 .  Inclusion/Exclusion criteria. Criteria for inclusion: 1) the study was a case-control or cohort study; 2) the study should use DTI technique to screen on substantianigra in both PD patients and healthy controls; 3) the study should have the detailed data on FA, and MD or ADC (MD and ADC were not entirely equal, but both indexes were the descriptors of total magnitude of diffusion and should all be included in the analysis); 4) for longitudinal trials, only the data from the first period was included. Review articles, comments, editorials, and case reports were all excluded in our study.
Data extraction. The following information was used in our meta-analysis, documented case by case: i) the authors, publication date, the number of participants, ages, disease duration, Unified Parkinson's Disease Rating Scale (UPDRS) motor scores, Hoehn-Yahr (H-Y) stage, and medications; ii) MRI scanner parameters, including acquisition characteristics, field strength of the MRI scanner, head coil, voxel size, and DTI directions; iii) the method of delineating the neuroanatomy of SN and the way of DTI processing the analyzing; iv) primary data value outcomes (FA, MD or ADC). If the original studies provided with only parameters like median, range or quartiles instead of mean and standard deviation, the conversion procedures were performed and estimated by previously published conversion techniques 31 to conform with the meta-analysis used previously 22 . Statistical analysis. The pooled weighted mean difference (WMD) and their 95% confidence intervals (CI) were calculated by the method of inverse variance with a random-effect or fixed-effect model. Statistical heterogeneity was evaluated using Higgins' I 2 -statistics, in which I 2 > 50% was considered heterogeneous among included studies. If I 2 < 50%, a fixed-effect model was applied further, otherwise, a random-effect model was used instead. Funnel plot symmetry and Egger's tests were used to evaluate the publication bias. Review Manager 5.3 and stataSE12.0 software were used to perform data analyses in current study. P value < 0.05 was considered statistically significant.

Disclosure.
We would like to declare that the work has not been published previously. On behalf of all authors, the corresponding author states that there is no conflict of interest.

Conclusion
This meta-analysis on measurements of FA in SN by DTI technique finds lower FA values in PD patients compared to the HCs. The strength of this finding indicates that FA measurement though DTI could be potentially used as a surrogate (complementary), non-invasive marker to diagnose PD more precisely. DTI could also be used to monitor the changes of both the brain structure and its functions for PD patients. Future research should focus on DTI measurement of image changes for early PD lesions, as well as its possible use as a preclinical diagnostic tool for those at risk of developing PD.