The “Central Vein Sign” on T2*-weighted Images as a Diagnostic Tool in Multiple Sclerosis: A Systematic Review and Meta-analysis using Individual Patient Data

We aimed to evaluate the pooled incidence of central vein sign on T2*-weighted images from patients with multiple sclerosis (MS), and to determine the diagnostic performance of this central vein sign for differentiating MS from other white matter lesions and provide an optimal cut-off value. A computerized systematic search of the literature in PUBMED and EMBASE was conducted up to December 14, 2018. Original articles investigating central vein sign on T2*-weighted images of patients with MS were selected. The pooled incidence was obtained using random-effects model. The pooled sensitivity and specificity were obtained using a bivariate random-effects model. An optimal cut-off value for the proportion of lesions with a central vein sign was calculated from those studies providing individual patient data. Twenty-one eligible articles covering 501 patients with MS were included. The pooled incidence of central vein sign at the level of individual lesion in patients with MS was 74% (95% CI, 65–82%). The pooled sensitivity and pooled specificity for the diagnostic performance of the central vein sign were 98% (95% CI, 92–100%) and 97% (95% CI, 91–99%), respectively. The area under the HSROC curve was 1.00 (95% CI, 0.99–1.00). The optimal cut-off value for the proportion of lesions with a central vein sign was found to be 45%. Although various T2*-weighted images have been used across studies, the current evidence supports the use of the central vein sign on T2*-weighted images to differentiate MS from other white matter lesions.

Studies were excluded if any of the following exclusion criteria were satisfied: (1) conference abstracts; (2) review articles; (3) case reports or case series including fewer than five patients; (4) letters, editorials, and short surveys; (5) studies with a partially overlapping patient cohort, and (6) animal studies. For studies with a partially overlapping study population, the study including the largest number of patients was selected. Authors of potentially eligible articles that did not provide sufficient information were contacted for the provision of further data.
Data extraction and quality assessment. The incidence of central vein sign on T2*-weighted images from patients with MS and the diagnostic performance of the central vein sign for differentiating MS from other white matter lesions were extracted from the eligible articles. Central vein sign on T2*-weighted imaging was defined as follows: (1) the vein should appear as a thin line or dot; (2) when technically possible, the vein should be visualized in at least two perpendicular planes; and (3) the vein can run partially or entirely through the lesion, but must be located centrally, regardless of the lesion's shape 5 . Two by two tables (true positive, false positive, false negative, true negative) for determination of the diagnostic performance of the central vein sign for differentiating MS from other white matter lesions such as small vessel disease, CNS inflammatory vasculopathies, or NMOSD were also constructed. If the diagnostic performances of multiple MRI sequences were separately evaluated, the results with the highest performance were selected. If a two by two table could not be acquired, the authors were contacted for provision of further data by E-mail.
The following information was extracted from the eligible studies: (1) the institution, the study period, study design (retrospective or prospective design), consecutive or non-consecutive patient enrollment, and the reference standard; (2) the number of MS patients, mean age, age range, and female to male ratio; (3) the magnetic field strength of the scanner, scanner manufacturer, scanner model, MRI sequence, and cut-off values for the proportion of lesions with central vein sign used to diagnose MS; and (4) the number of MRI readers, and blindness to the reference standard.
Quality assessment was performed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) criteria 29 . The literature search, study selection, data extraction, and quality assessment were performed by two reviewers (C.H.S., S.J.K.).

Statistical analyses.
The pooled incidence of central vein sign on T2*-weighted images of MS was calculated with the inverse variance method for calculating weights and the DerSimonian-Liard random-effects model [30][31][32] . Heterogeneity was assessed by Higgins inconsistency index (I 2 ) test, with values greater than 50% taken as indicating substantial heterogeneity 33 . Publication bias was assessed by a funnel plot, and the statistical significance was assessed by Egger's test 34 . Meta-regression was conducted to explain the effects of study heterogeneity. The following covariates were considered: (1) study design (prospective study vs. other); (2) MRI sequence (studies including FLAIR* [combined FLAIR and T2*-weighted images] 35 vs. other); (3) reader (radiologist vs. other); (4) reader blindness to the reference standard; and (5) patient age (age ≤41 [median value] vs. age >41). Subgroup analyses according to the strength of the MRI scanner (7, 3, and 1.5-Tesla) were also performed.
The pooled sensitivity and specificity and their 95% confidence intervals [CI] for the diagnostic performance of central vein sign on T2*-weighted images for differentiating MS from other white matter lesions were calculated using a bivariate random-effects model [30][31][32] . A coupled forest plot of sensitivity and specificity and a hierarchical summary receiver operating characteristic (HSROC) curve with 95% confidence and prediction regions were plotted. Heterogeneity was assessed by the following methods: (1) Cochran's Q-test (p < 0.05 indicating the presence of heterogeneity); (2) Higgins I 2 test (a value >50% indicating the presence of heterogeneity); 33 (3) visual assessment of the coupled forest plot for the presence of a threshold effect, i.e., a positive correlation between sensitivity and false positive rate; and (4) the Spearman correlation coefficient between sensitivity and false positive rate (a value >0.6 indicating a threshold effect) 36 . Publication bias was assessed by Deeks' funnel plot, with the statistical significance being assessed by Deeks' asymmetry test 37 . A meta-regression was conducted to explain the effects of study heterogeneity, with the following covariates being utilized for the bivariate meta-regression model: (1) study design (prospective study vs. other); (2) MRI sequence (studies including FLAIR* 35 vs. other); (3) reader (radiologist vs. other); (4) reader blindness to the reference standard; and (5) patient age (age ≤41 [median value] vs. age >41). Subgroup analysis was conducted on those studies using a proportion of lesions with central vein sign as a cut-off value.
An optimal cut-off value for the proportion of lesions with central vein sign was calculated from those studies providing individual patient data. The individual patient data were extracted from the articles, and when not reported, Plot Digitizer 2.6.8 (plotdigitizer.sourceforge.net) was used to estimate the data from plots indicating the proportion of lesions with central vein sign. The sensitivity and specificity of the central vein sign and the corresponding cut-off value for the proportion of lesions with a central vein sign were estimated using the Youden index. The Youden index is defined as sensitivity + specificity -1, with it having a minimum value of −1 and a maximum value of + 1, with a value of + 1 indicating the optimal value for an algorithm 38 .
Statistical analyses were performed by one reviewer (C.H.S., with 6 years of experience in performing systematic reviews and meta-analysis) using the "metafor" and "mada" packages in R v.3.4.1 (R Foundation for Statistical Computing, Austria), and the "metandi" and "midas" modules in STATA 15.0 (StataCorp, College Station, USA).

Results
Literature search. The details of the study selection process are illustrated in Fig. 1 and Supplementary materials. Finally, 21 eligible articles encompassing 501 patients with MS were included in the analyses 6-26 . Characteristics of the included studies. The characteristics of the eligible studies are shown in Table 1.
Quality assessment. The quality of the 21 eligible studies was considered as moderate, with more than four of the seven domains being satisfied ( Supplementary Fig. 1). The details of the quality assessment are described in Supplementary materials.  www.nature.com/scientificreports www.nature.com/scientificreports/ from 40% to 92%, while the pooled incidence of central vein sign on T2*-weighted images was 74% (95% CI, 65-82%; Fig. 2). Heterogeneity was present among these values (I 2 = 98%). A meta-regression was performed to explore the effects of heterogeneity, and among the various covariates analyzed, the study design showed statistical significance (p = 0.01). Prospective studies showed a significantly higher pooled incidence of central vein sign on T2*-weighted images (86%; 95% CI, 80-91%) than retrospective studies (67%; 95% CI, 56-77%). Other covariates including MRI sequence (p = 0.08), reader (p = 0.51), reader blindness to the reference standard (p = 0.93), and age (p = 0.31) did not show statistically significant differences. There was no publication bias (p = 0.63; Supplementary Fig. 2).

Diagnostic performance of the central vein sign on T2*-weighted Images for diagnosis of MS.
Twelve original articles evaluated the overall diagnostic performance of the central vein sign on T2*-weighted images for differentiating MS from other white matter lesions [6][7][8][9]12,14,15,20,21,23,24,26 . Four studies included patients with small vessel disease as a comparison group 6,7,21,26 , two studies included patients with NMOSD 8,23 , two studies included patients with CNS inflammatory vasculopathies 9,20 , two studies included healthy controls 12,14 , one study included non-MS white matter lesions 15 , and one study included patients with migraine who had been erroneously diagnosed with MS 24 . www.nature.com/scientificreports www.nature.com/scientificreports/ Ten of the twelve studies used a cut-off parameter based on the proportion of lesions with central vein sign on T2*-weighted images [6][7][8][9]12,14,20,21,23,26 , and the patients with MS showed significantly higher proportions of lesions with central vein sign than did the patients with other white matter lesions. One study used just the presence of a central vein sign 15 , and one study used a simplified three-lesion algorithm 24 . The individual sensitivities and specificities both varied from 80% to 100%. The pooled sensitivity was 98% (95% CI, 92-100%), and the pooled specificity was 97% (95% CI, 91-99%; Fig. 3). The area under the HSROC curve was 1.00 (95% CI, 0.99-1.00; Fig. 4).
Both the Q-test (Q = 2.636, p = 0.13) and the Higgins I 2 statistic (I 2 = 24%) demonstrated that the possibility of heterogeneity was low across the studies. The coupled forest plot revealed no evidence of a threshold effect (Fig. 3), and the Spearman correlation coefficient was −0.092 (95% CI, −0.632-0.509), also indicating no threshold effect. The Deeks' funnel plot demonstrated that publication bias was present (p < 0.01; Supplementary Fig. 3).
The individual cut-off values ranged from 30% to 67%, with a median value of 45%. The area under the ROC curve of the proportion of lesions with central vein sign for the diagnosis of MS was 0.994 (95% CI, 0.975-1.000; Fig. 5). The optimal cut-off value was 45% using the Youden index, resulting in a sensitivity of 97% (95% CI, 94-99%) and specificity of 99% (95% CI, 92-100%).  Table 2. MRI characteristics of the eligible studies. SWI = susceptibility-weighted imaging, NA = not available, SWAN = susceptibility-weighted angiography.

Discussion
The current study revealed a high incidence (74%) of central vein sign on T2*-weighted images of patients with MS, and also revealed that the central vein sign has excellent diagnostic performance for differentiating MS from other white matter lesions, with a pooled sensitivity of 98% and a pooled specificity of 97%. Using individual patient data, the optimal cut-off value for the proportion of lesions with central vein sign on T2*-weighted images was found to be 45%. Although various T2*-weighted images have been used across studies, the current evidence supports the use of the central vein sign on T2*-weighted images to differentiate MS from other white matter lesions.
The differentiation of MS from other white matter lesions can sometimes be challenging, both clinically and radiologically. The proportion of lesions exhibiting the central vein sign is thought to be useful for differentiating MS from some of its mimics 1 . Our results also showed excellent diagnostic performance for differentiating MS from other white matter lesions according to the proportion of lesions exhibiting the central vein sign. In terms of pathophysiology, the inflammatory demyelination in MS spreads in the parenchyma with perivenular extension 41 . However, cerebral small vessel disease is thought to contribute to the chronic ischemic damage presenting at the arteriole 42 , and inflammatory vasculopathies affect medium and small arteries, and are characterized by inflammatory infiltrates of the vessel wall, fibrinoid necrosis, and thrombosis with ischemic change 43 . As the central vein sign is based on a pathological background, the central vein sign may become a promising biomarker for differentiating MS from other white matter lesions.
The current study highlights the fact that the determination of an optimal cut-off value for the proportion of lesions with a central vein sign on T2*-weighted images is clinically and radiologically important if standardized T2*-weighted images are to be used in daily clinical practice. We found that individual cut-off values ranged from 30% to 67%, and that the optimal cut-off value using individual patient data was 45%, resulting in a sensitivity of 97% and specificity of 99%. Although our results were outstanding, the application of this optimal cut-off value requires time-consuming lesion counting and frequency estimation, which may be difficult to conduct in daily clinical practice. A recent study showed the possibility of a fully automated method for detecting the central vein sign, demonstrating a promising performance 44 . However, further studies are needed to validate fully automated methods for detecting the central vein sign.
The North American Imaging in Multiple Sclerosis Cooperative mentioned that imaging of veins in the brain can be performed using T2*-based MRI sequences at any magnetic field strength (1.5, 3, or 7-Tesla) 5 . In addition, high-resolution isotropic T2*-weighted 3D EPI is currently the most promising sequence, and FLAIR* has the potential to become a standard clinical protocol 5 . However, these sequences have not been widely used because of the difficulty in optimizing protocols and post-processing. Therefore, standardization of T2*-weighted imaging is crucial. We found that five studies using FLAIR* demonstrated excellent diagnostic performance for diagnosing MS 7,10,12,20,24 . To generate FLAIR* images, co-registration, interpolation, and multiplication processes are needed 35 . For widespread dissemination of FLAIR*, manufacturer-provided software for direct automated image post-processing on the scanner is necessary.
Although our study results showed the area under the HSROC curve of 0.99 for diagnosing MS using the central vein sign, there are several issues should be considered. Our study is vulnerable to inclusion bias because of the selection of controls. Various comparison groups such as small vessel disease, NMOSD, CNS inflammatory vasculopathies, healthy controls, and non-MS white matter lesions were included. In addition, a previous study showed that the specificity for diagnosing MS using brain MRI with American Academy of Neurology criteria was only 29%, which indicated an increased risk of false-positive diagnosis of MS 45 . Therefore, careful clinical application should be made using our results in daily clinical practice.