Added Value of Vessel Wall Magnetic Resonance Imaging for Differentiation of Nonocclusive Intracranial Vasculopathies
Background and Purpose—Our goal is to determine the added value of intracranial vessel wall magnetic resonance imaging (IVWI) in differentiating nonocclusive vasculopathies compared with luminal imaging alone.
Methods—We retrospectively reviewed images from patients with both luminal and IVWI to identify cases with clinically defined intracranial vasculopathies: atherosclerosis (intracranial atherosclerotic disease), reversible cerebral vasoconstriction syndrome, and inflammatory vasculopathy. Two neuroradiologists blinded to clinical data reviewed the luminal imaging of defined luminal stenoses/irregularities and evaluated the pattern of involvement to make a presumed diagnosis with diagnostic confidence. Six weeks later, the 2 raters rereviewed the luminal imaging in addition to IVWI for the pattern of wall involvement, presence and pattern of postcontrast enhancement, and presumed diagnosis and confidence. Analysis was performed on per-lesion and per-patient bases.
Results—Thirty intracranial atherosclerotic disease, 12 inflammatory vasculopathies, and 12 reversible cerebral vasoconstriction syndrome patients with 201 lesions (90 intracranial atherosclerotic disease, 64 reversible cerebral vasoconstriction syndrome, and 47 inflammatory vasculopathy lesions) were included. For both per-lesion and per-patient analyses, there was significant diagnostic accuracy improvement with luminal imaging+IVWI when compared with luminal imaging alone (per-lesion: 88.8% versus 36.1%; P<0.001 and per-patient: 96.3% versus 43.5%; P<0.001, respectively). There was substantial interrater diagnostic agreement for luminal imaging+IVWI (κ=0.72) and only slight agreement for luminal imaging (κ=0.04). Although there was a significant correlation for both luminal and IVWI pattern of wall involvement with diagnosis, there was a stronger correlation for IVWI finding of lesion eccentricity and intracranial atherosclerotic disease diagnosis than for luminal imaging (κ=0.69 versus 0.18; P<0.001).
Conclusions—IVWI can significantly improve the differentiation of nonocclusive intracranial vasculopathies when combined with traditional luminal imaging modalities.
Early diagnosis of intracranial vasculopathies, including intracranial atherosclerotic disease (ICAD), reversible cerebral vasoconstriction syndrome (RCVS), and infectious/inflammatory vasculopathies (IVas), is important because inappropriate or delayed therapy can lead to worse outcomes.1–3 Current diagnostic imaging algorithms rely on luminal imaging for disease differentiation, with digital subtraction angiography (DSA) serving as the imaging gold standard.4 DSA, however, shows limited sensitivity/specificity that can be as low as 30% for IVas, and only 25% to 43% of pathologically proven primary angiitis have luminal abnormalities on angiography.4–8 If there is suspicion for IVas, invasive tests such as brain biopsy may be implemented, which not only carry a significant risk of morbidity but also show sensitivities of only 53% to 80% for vasculitis.9,10 ICAD frequently remodels outwardly, resulting in luminal-based underestimation of disease burden,11 and luminal imaging may not detect culprit plaques at all.12 DSA also serves as the imaging standard for RCVS; however, the imaging appearance of arterial beading is nonspecific and indistinguishable from IVas.9,13
Intracranial vessel wall magnetic resonance imaging (MRI) has shown promise in its ability to differentiate and characterize intracranial vasculopathies.14 The inclusion of intracranial vessel wall MRI (IVWI) can better differentiate between moyamoya disease and causes of moyamoya syndrome compared with luminal imaging alone.15 Causes of nonocclusive intracranial vasculopathies can be differentiated using a multicontrast IVWI protocol.16 However, it is unknown whether IVWI has added benefit in evaluating nonocclusive intracranial arteriopathies compared with luminal imaging alone. This study compares the diagnostic accuracy of IVWI+luminal imaging compared with luminal imaging alone in nonocclusive vasculopathy differentiation, specifically ICAD, RCVS, and IVas.
The authors will make data available for study replication on request.
After institutional review board study approval with waiver of consent, consecutive patients with arterial wall imaging from December 2012 through February 2016 were reviewed from a prospectively maintained database. We extracted cases with documented luminal irregularity/narrowing but without occlusion on the clinically acquired computed tomographic angiography, magnetic resonance angiography, and DSA. Two stroke neurologists (K.J.B. and A.D.H.) reviewed the clinical data and luminal imaging reports, while blinded to patient identifiers, IVWI information, and clinical diagnosis, and categorized the vasculopathies as ICAD, RCVS, and IVas on a per-patient basis. In terms of final diagnosis, cases diagnosed as RCVS were required to show reversibility of stenosis on follow-up luminal imaging within 3 months. For the diagnosis of IVas, there was a requirement for histological or cerebrospinal fluid evidence of infection/inflammation in combination with arterial stenosis and MRI brain findings compatible with IVas. ICAD diagnosis could not have evidence of central nervous system inflammation or short-term reversibility of arterial lesions. If there was disagreement in the diagnosis, a third stroke neurologist (D.L.T.) arbitrated. The neurologist review served as the diagnostic gold standard. Cases with other diagnoses, multiple diagnoses, or a diagnosis could not be agreed on by the neurology reviewers were excluded.
Patients were scanned on a 3T Siemens Trio MR scanner (Siemens Healthcare, Erlangen, Germany) using a standard head coil. The IVWI protocol included high-resolution multiplanar T1-weighted (0.4×0.35-mm in-plane resolution; 2-mm slice thickness; time to repetition/time to echo, 1000/10 ms; and 36 seconds per slice) pre- and post-contrast, T2-weighted (0.4×0.4-mm in-plane resolution; 1-mm slice thickness; time to repetition/time to echo, 3550/72 ms; and 9.3 seconds per slice), and 3D SPACE T2-weighted (0.6×0.6-mm in-plane resolution; 0.6-mm slice thickness; time to repetition/time to echo, 2400/80 ms; 64 slices; and 10:20 minutes) sequences. More detailed imaging parameters can be found in previous publications.15,16
A single rater (M.M.-B.) reviewed luminal imaging of the included cases independent of diagnosis, clinical data, and IVWI to determine the arterial segments with luminal irregularity/stenosis. Luminal imaging was used to identify lesions in order to avoid potential artifacts from nonsuppression of blood flow or cerebrospinal fluid mimicking lesions on IVWI. The abnormal arterial segments were recorded and used to guide the raters to the lesions to evaluate. Two other raters (D.K.S. and D.K.H.) underwent training in IVWI interpretation, which included review of a packet of articles relating to IVWI interpretation, in-person case reviews, didactic lecture, and independent test case review with feedback provided. IVWI characteristics and criteria for each vasculopathy evaluated are included in Table I in the online-only Data Supplement. The 2 independent raters blinded to clinical and IVWI data reviewed consecutive luminal imaging studies (computed tomographic angiography, magnetic resonance angiography, and DSA performed for each subject prior to IVWI) for the vasculopathy subjects. The raters evaluated each lesion individually and independent from other lesions but with knowledge of what additional arterial segments were affected by lesions for each subject (to provide an idea for global disease involvement). Lesions were randomized so each lesion could be reviewed based on its imaging characteristics independent of other patient lesions. The raters evaluated luminal imaging for the pattern of involvement (concentric/eccentric), diagnosis (ICAD, IVas, RCVS, or unknown if diagnosis could not be narrowed down to a single disease), and confidence in their diagnosis on a 4-point Likert scale (0=50% confidence, 1=51% to 75% confidence, 2=76% to 90% confidence, and 3=>90% confidence).15 On luminal imaging, eccentric lesions were defined as those primarily affecting one side of the lumen on 2-dimensional review or ≤3 sides of the lumen on 3-dimensional review, whereas concentric was defined as lesions with circumferential narrowing of the lumen. After a 6-week washout period, the raters reevaluated the randomized lesions using both luminal imaging+IVWI individually and independently. The raters evaluated the pattern of wall involvement (eccentric/concentric), presence of enhancement (yes/no), pattern of enhancement (focal, heterogeneous, and diffuse), diagnosis, and confidence in diagnosis (the 4-point scale).15 Concentric lesions were defined as those where the width of the thinnest wall segment was >50% that of the thickest segment, whereas for eccentric lesions, the thinnest segment had a width <50% that of the thickest segment.17 For the enhancement pattern, focal was a punctate or short linear focus of enhancement, heterogeneous was incomplete enhancement, and diffuse was complete enhancement.15,16
Continuous and categorical variables were summarized as mean±SD and count (percentage), respectively. Clinical characteristics of each subject were summarized and compared between the 3 clinical diagnoses using the Kruskal–Wallis test (continuous variables) and Fisher exact test (categorical variables). For most of the analysis, the units of analysis were the lesion or the read of the lesion, where the read is the assessment of a lesion by a single rater. Thus, there were twice as many reads as lesions.
Throughout, analyses were conducted with both rater’s reads pooled together, which produces results that correspond to the average of the 2 raters and generally lead to an increase in statistical power. Analyses were repeated for each rater separately to determine whether there were any material differences in the rater-specific results. Multiple lesions per subject and multiple raters per lesion were not treated as independent observations, however. Permutation tests with resampling performed by subject (all lesions and reads from the same subject were permutated together, maintaining their dependence)18 or generalized estimating equations models19 clustered by subject were used to account for dependence among the lesions and reads from the same subject when conducting hypothesis tests. Confidence intervals were calculated using the percentile method of the nonparametric bootstrap with resampling by subject.18
Interrater agreement was summarized as percent agreement (100%×number of lesions where both raters gave the same rating divided by the total number of lesions) and Cohen κ. Percent agreement and Cohen κ were compared between the luminal imaging only session and the luminal imaging+IVWI session using the nonparametric bootstrap with resampling by subject as described above.
Diagnostic accuracy was assessed on per-lesion and per-patient bases, computed as the percentage of reads where the rater’s diagnosis matched the final diagnosis. Under the per-lesion analysis, any reads classified as uncertain or equivocal by the rater, without a rater diagnosis, were considered an incorrect diagnosis. Under the per-patient analysis, the per-lesion reads were aggregated to per-patient reads by first summing the confidence scores of each lesion for each vasculopathy diagnosis (ICAD, RCVS, and IVas) separately and then selecting the per-patient diagnosis as the one with the highest total confidence score. The per-patient diagnosis was considered uncertain/equivocal if ≥50% of the constituent lesions had uncertain/equivocal diagnoses or multiple vasculopathy diagnoses were tied in total confidence score. An uncertain/equivocal per-patient diagnosis was considered incorrect for calculating diagnostic accuracy, as with the per-lesion analysis. Diagnostic accuracy was compared between the luminal imaging and the luminal imaging+IVWI sessions using a permutation test based on McNemar test, resampled by subject.
Diagnostic confidence, dichotomized as highly confident (rating=3) versus not highly confident (rating <3), was compared between the luminal imaging and luminal imaging+IVWI sessions using generalized estimating equations–based logistic regression models. These comparisons were performed using all reads and within the subgroups where the rater diagnosis was concordant with the final diagnosis and where the rater diagnosis was discordant with the final diagnosis. Individual luminal imaging and IVWI findings were compared between the final diagnosis groups using permutation tests based on the χ2 test, resampled by subject. All statistical calculations were conducted with the statistical computing language R (version 3.1.1; R Foundation for Statistical Computing, Vienna, Austria). Throughout, 2-tailed tests were used with statistical significance defined as P<0.05.
Clinical Diagnoses and Luminal Characteristics
Two hundred and twenty consecutive IVWI cases with luminal imaging were reviewed, of which 54 patients were included in this study: 30 ICAD, 12 RCVS, and 12 IVas cases. IVas cases consisted of 4 varicella vasculopathies, 3 primary angiitis, 2 bacterial vasculopathies, and 1 each of Behcet-associated vasculopathy not otherwise specified, tuberculosis, and fungal vasculopathies. Patient clinical and demographic information is in Table II in the online-only Data Supplement. There were significant differences for number of vascular risk factors between vasculopathies because 97% of ICAD, 8% of RCVS, and 0% of IVas subjects had ≥2 vascular risk factors, respectively (P<0.001). Luminal imaging performed in each vasculopathy is also listed in Table II in the online-only Data Supplement. Magnetic resonance angiography was performed on all patients. There were variations in the frequency of DSA (25% to 67%; P=0.080) and computed tomographic angiography (57% to 75%; P=0.41) performance between disease groups, although these differences were not statistically significant. All patients underwent imaging evaluation based on clinical practice and need.
From these 54 patients, 201 lesions (90 ICAD, 64 RCVS, and 47 IVas lesions) were assessed by the 2 raters, for a total of 402 ratings. There were 1 to 10 lesions per patient (median 3). For a description of arterial segments involved for each disease, please refer to Table III in the online-only Data Supplement. There was a significant overall association between arterial segment involvement and final diagnosis (P=0.007), although this was because of a significant difference in intracranial internal carotid artery involvement (P=0.007) without a significant difference in disease involvement between other arterial segments (P=0.28). ICAD was significantly more likely to involve the internal carotid artery (42.2%) than IVas (9.4%; P=0.046) and RCVS (25.5%; P=0.001).
Vessel Wall MRI Characteristics
There were significant differences in the IVWI characteristics between ICAD, RCVS, and IVas (Table 1). Typical findings are shown in Figure 1. On IVWI, ICAD lesions were most commonly eccentric (91.1%) and diffusely or heterogeneously enhancing (86.7%). RCVS typically had concentric lesions (80.5%) that showed no enhancement (53.9%) or diffuse enhancement (32.8%; Figure 2). IVas most commonly showed concentric (76.6%), diffusely enhancing lesions (81.9%).
Pattern of involvement by both luminal and IVWI imaging was significantly associated with IVWI disease states (Table 1), specifically ICAD lesions were more likely to be eccentric than RCVS and IVas lesions. However, there was a significantly stronger correlation between final ICAD diagnosis and lesion eccentricity on IVWI than lesion eccentricity detected on luminal imaging alone (κ=0.69 versus 0.18; P<0.001).
Added Benefit of Vessel Wall MRI
On the basis of luminal imaging alone, raters made the correct per-lesion diagnosis in 145 of 402 evaluations (36.1%). When luminal imaging+IVWI were reviewed, rater accuracy significantly increased with 357 out of 402 (88.8%) evaluations correctly diagnosed (P<0.001; Table 2). The increase in diagnostic accuracy was significant for each rater individually (rater 1: 26.4% versus 83.6%; P<0.001 and rater 2: 45.8% versus 94.0%; P<0.001). Diagnostic accuracy significantly improved with the addition of IVWI for each disease group (Table 2). Of the 257 reads with an incorrect diagnosis on luminal imaging, 221 (86.0%) were correctly reclassified with the addition of IVWI. Of the 145 reads with the correct diagnosis on luminal imaging, 9 (6.2%) were incorrectly reclassified with the addition of IVWI (5 ICAD misclassified as IVas, 2 ICAD to RCVS, 2 IVas to RCVS). Thirty-six reads (8.9% of all 402 reads) had an incorrect diagnosis on both evaluations (50% RCVS, 25% ICAD, and 25% IVas).
The per-patient analysis involved 108 evaluations (54 patients×2 raters). Using luminal imaging alone, raters made the correct diagnosis at the patient level in 47 out of 108 (43.5%) evaluations (Table 2). Per-patient diagnostic accuracy increased to 104 out of 108 (96.3%) when luminal+IVWI were reviewed compared with luminal imaging alone (P<0.001). Similar to the per-lesion analysis, improvement in per-patient diagnostic accuracy was most apparent for RCVS (0.0%–100.0%; P<0.001) and IVas (8.3% versus 95.8%; P<0.001), although there was also improvement in diagnostic accuracy of ICAD (75.0%–95.0%; P=0.001).
There was a significant increase in confidence between luminal imaging alone compared with luminal imaging+IVWI (high confidence rating of 3: 17% versus 44%; P<0.001). There was a significant increase in confidence with the inclusion of IVWI when rater diagnosis matched the final diagnosis (34.5% versus 47.9%; P=0.002). However, there was no significant change in confidence ratings when the rater diagnosis did not match the final diagnosis (7.8% versus 13.3%; P=0.15).
Interrater agreement on luminal characteristics and diagnosis for luminal and IVWI is summarized in Table 3. There was only slight agreement for pattern of involvement of luminal imaging. Interrater agreement for diagnosis on luminal imaging was slight (κ=0.04; 95% confidence interval, −0.03 to 0.12) and substantial for IVWI (κ=0.72; 95% confidence interval, 0.64–0.8). This difference in agreement was significant (P<0.001).
We report the first study to assess the added benefit of IVWI compared with current luminal diagnostic algorithms in the differentiation of nonocclusive vasculopathies, specifically ICAD, IVas, and RCVS. This study shows that inclusion of IVWI can significantly improve imaging diagnostic differentiation of these vasculopathies compared with luminal imaging alone. In previous studies, we established substantial to almost perfect agreement in IVWI characteristics for nonocclusive16 and steno-occlusive15 intracranial vasculopathy differentiation. There was only slight agreement in pattern of involvement on luminal imaging in the current study, indicating that this is a less reliable and reproducible assessment tool. This is especially important considering that luminal imaging is the standard of care for intracranial vasculopathy evaluation and differentiation. In addition, the arterial segment of disease involvement showed no significant correlation with final diagnosis when intracranial internal carotid artery involvement was excluded. The likelihood of a correct diagnosis in the setting of nonocclusive vasculopathy significantly increased when IVWI was evaluated in addition to luminal imaging (per-lesion: 36.1%–88.8% and per-patient: 43.5%–96.3%), and this increase was significant overall and for ICAD, RCVS, and IVas individually. There was substantial interrater diagnostic agreement for IVWI+luminal imaging, whereas only slight agreement for luminal imaging assessment alone.
Historically, angiographic imaging has served as the reference standard for the differentiation and characterization of nonocclusive vasculopathies. Specifically, DSA is considered the imaging gold standard, with differentiating features for ICAD from RCVS and IVas being lesion location and pattern of involvement. In the current study, however, 12% of ICAD lesions involved distal branches, whereas 41% of RCVS and 32% of IVas involved first-order branches of the anterior, middle, and posterior cerebral arteries, indicating prominent overlap in traditional patterns of disease involvement. In terms of pattern of involvement, there was a significant difference in eccentricity on luminal imaging of ICAD lesions relative to RCVS and IVas; however, 53.2% of RCVS and 51.1% of IVas lesions were rated as eccentric on luminal imaging as compared with 71.2% of ICAD lesions. In addition, there was a stronger correlation between ICAD diagnosis and the described eccentric pattern on IVWI than there was for luminal imaging. The current study, similar to previous studies,16,17,20,21 establishes distinctive IVWI patterns for ICAD, RCVS, and IVas that can improve diagnostic accuracy compared with luminal imaging alone. ICAD typically showed eccentric, heterogeneously or diffusely enhancing lesions; IVas showed concentric, diffusely enhancing lesions; and RCVS had concentric nonenhancing or diffusely enhancing lesions.
A few previous studies have compared the IVWI appearances of ICAD, RCVS, and IVas. With a multicontrast IVWI protocol,16 there were significant differences in IVWI appearance of these vasculopathies, with the following typical descriptions for each disease—ICAD: eccentric, diffuse or incompletely, mildly or moderately enhancing lesion with mixed T2 signal intensity; RCVS: concentric, nonenhancing (when enhancement was present, it was diffuse and mild) lesion with minimal iso- or hypo-intense wall thickening; and IVas: concentric, diffusely and moderately enhancing lesions. Obusez et al21 compared 13 IVas and 13 RCVS cases with IVWI and found that 9 out of 13 IVas cases had smooth, concentric, strong (defined as thick-walled enhancement) enhancement; 3 out of 13 showed eccentric, strong involvement; and 1 case showed no enhancement. RCVS showed smooth concentric wall thickening in 10 out of 13 cases, 4 of which showed mild enhancement (defined as thin-walled enhancement), whereas 3 had no wall abnormality. These findings are similar to our study, although we found a higher occurrence of vessel wall enhancement in RCVS (46.1% compared with 18.2%16 and 31%21). Other studies have also documented the presence of enhancement in cases of RCVS.22 These studies, however, established the intensity of enhancement or the presence of wall thickening, if present, to be a differentiating characteristic for enhancing RCVS (thin-walled/mild enhancement) compared with IVas (thick-walled/moderate enhancement).16,20,21 Mandell et al20 compared 3 cases of RCVS with 4 cases of IVas and found RCVS to have minimal or absent enhancement, whereas all IVas cases showed wall enhancement. Swartz et al17 compared 13 ICAD and 3 IVas cases and found that ICAD showed eccentric enhancing lesions, whereas IVas typically had concentric enhancing lesions.
There are several limitations of this study. This was a retrospective imaging review. Luminal imaging modalities performed were heterogeneous; however, best practice guidelines were used for imaging and patient care and the image review approximates clinical practice at many institutions. Histological confirmation was not available for most cases. The number of RCVS and IVas cases in this study is limited, which are relatively rare conditions. There is the possibility that some patients may have had lesions from more than one vasculopathy; however, we attempted to limit this by excluding patients with more than one suspected diagnosis. This study only evaluates the added value of IVWI in differentiating certain vasculopathies because only a subset of potential vascular diseases is studied. No healthy reference is provided for comparison to vascular pathological states. Lesions were evaluated individually and independently. Although this does not match typical clinical review for luminal imaging, each lesion should be reviewed individually on IVWI. To mitigate this discrepancy, the raters were aware of additional lesions per patient during lesion review. We used 2D T1 turbo-spin echo IVWI sequence before and after contrast, which requires many planes of scanning and requires more scan time and physician supervision compared with isotropic 3D IVWI acquisitions, which can be reformatted into multiple planes with a single acquisition. This was a single-center study performed on a single MRI platform. For these reasons, a large, multicenter, multiplatform prospective study is needed to confirm these findings.
The addition of IVWI to the diagnostic algorithm improves diagnostic accuracy and confidence in the differentiation of nonocclusive intracranial vasculopathies, specifically ICAD from RCVS and IVas when compared with luminal imaging alone. This differentiation is important because treatment algorithms for each condition differ significantly. With further confirmation of the added benefit of IVWI in disease differentiation, its inclusion in diagnostic algorithms may improve patient management and outcomes while potentially limiting invasive diagnostic tests.
Sources of Funding
This study was funded by the National Institutes of Health grants R56 NS092207 01 and R01 NS092207 01A1.
D.S. Hippe receives grants from GE, Philips, and Toshiba Healthcare unrelated to the current work. Dr Hatsukami and Dr Yuan receives grants from Philips Healthcare unrelated to the current work. The other authors report no conflicts.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.117.018227/-/DC1.
- Received May 29, 2017.
- Revision received August 8, 2017.
- Accepted September 18, 2017.
- © 2017 American Heart Association, Inc.
- Kimmoun A,
- Baux E,
- Das V,
- Terzi N,
- Talec P,
- Asfar P,
- et al
- Singhal AB,
- Topcuoglu MA
- Calabrese LH,
- Furlan AJ,
- Gragg LA,
- Ropos TJ
- Bhattacharyya S,
- Berkowitz AL
- Torres J,
- Loomis C,
- Cucchiara B,
- Smith M,
- Messé S
- Mazighi M,
- Labreuche J,
- Gongora-Rivera F,
- Duyckaerts C,
- Hauw JJ,
- Amarenco P
- Miller TR,
- Shivashankar R,
- Mossa-Basha M,
- Gandhi D
- Mossa-Basha M,
- Alexander M,
- Gaddikeri S,
- Yuan C,
- Gandhi D
- Mossa-Basha M,
- de Havenon A,
- Becker KJ,
- Hallam DK,
- Levitt MR,
- Cohen WA,
- et al
- Mossa-Basha M,
- Hwang WD,
- De Havenon A,
- Hippe D,
- Balu N,
- Becker KJ,
- et al
- Davison AC,
- Hinkley DV
- Diggle P HP,
- Liang KY,
- Zeger S
- Mandell DM,
- Matouk CC,
- Farb RI,
- Krings T,
- Agid R,
- terBrugge K,
- et al
- Obusez EC,
- Hui F,
- Hajj-Ali RA,
- Cerejo R,
- Calabrese LH,
- Hammad T,
- et al