Imaging features of adult moyamoya disease patients with anterior intracerebral hemorrhage based on high-resolution magnetic resonance imaging

Abstract

This study aimed to identify the high-resolution magnetic resonance imaging (HRMRI) features of moyamoya disease (MMD) patients with anterior intracerebral hemorrhage (ICH) and attempted to reveal potential mechanisms of anterior ICH. Eligible adult MMD patients were consecutively included, and the morphological features of lenticulostriate arteries (LSAs), vessel wall structure of terminal internal carotid artery (ICA) and periventricular anastomosis were evaluated by HRMRI. 78 MMD patients containing 21 patients with anterior ICH, 31 ischemic patients and 26 asymptomatic patients were included. The mean value of total length of LSAs in anterior ICH group (90.79 ± 37.00 mm) was distinctively lower (p < 0.001) compared with either ischemic group (138.04 ± 46.01 mm) or asymptomatic group (170.50 ± 39.18 mm). Lumen area of terminal ICA was significantly larger (p < 0.001) in hemorrhagic group (4.33 ± 2.02 mm²) compared with ischemic group (2.29 ± 1.17 mm²) or asymptomatic group (3.00 ± 1.34 mm²). Multivariate analysis revealed the total length of LSAs (OR 0.689, 95%CI, 0.565–0.840; p < 0.001) and lumen area of terminal ICA (OR 2.085, 95%, 1.214–3.582; p = 0.008) were significantly associated with anterior ICH. Coexistence of reduced LSAs and relatively preserved lumen area of terminal ICA with an AUC of 0.901 (95%CI, 0.812–0.990) could be a potential predictor of anterior ICH in MMD patients.

Introduction

Moyamoya disease (MMD) is a chronic cerebrovascular disease characterized by progressive stenosis and occlusion at Willis circle with compensatory development of a hazy collateral network at the base of the brain. ¹ Intracranial hemorrhage is the most severe event of MMD patients resulting in poor prognosis. ² The prevalence of intracranial hemorrhage was reported up to be 50% in East Asian patients,^3,4 and to be about 10%–30% in Caucasions.^5–7 Given the poor prognosis, it’s pivotal to clarify the mechanism of cerebral bleeding due to moyamoya disease.

Due to the complicated collateral channels induced by moyamoya disease, different hemorrhagic patterns may have various pathways and risk factors. ⁸ The sources of posterior hemorrhage of MMD patients are thought to be dilated choroidal and thalamic collaterals, which has been revealed in Japan Adult Moyamoya (JAM) Trial. ⁹ Nevertheless, the mechanisms of anterior hemorrhage are still not clarified.^8,10 Lenticulostriate arteries (LSAs) were considered possible sources of anterior intracerebral hemorrhage (ICH) according to the anatomical distribution of arteries in anterior area like putamen, caudate head, claustrum. However, previous studies failed to find the relationship between LSAs profiles and intracranial hemorrhage in MMD patients.^11–14 The LSAs profiles explored in previous studies were the proliferating capillaries or anastomotic networks connecting LSAs and cortical vessels (periventricular anastomosis) evaluated by digital subtraction angiography (DSA). Nevertheless, quantitative assessments of LSAs could be arduous in DSA images attributed to the overlapping vessels, unclean background and dynamic flow alterations, and measurements issues. ¹⁵ In addition, lack of anatomic landmarks also leads to faint identification of LSAs.

Recently, techniques of magnetic resonance imaging (MRI) to clearly exhibited LSAs have emerged gradually.^15–17 High resolution magnetic resonance imaging (HRMRI) has become a validated technique for assessment of both vessel wall structure and morphology of LSAs with comparable image quality to that of 7T TOF-MRA in healthy volunteers and patients with intracranial atherosclerotic disease.^16–19 Two previous studies applied this technique in moyamoya disease patients, indicating that LSAs of MMD patients could be clearly distinguished from other perforating arteries.^20,21 Thus, HRMRI was performed in our study to distinguish LSAs from overlapping vessels and further explore the morphologic changes of LSAs, vessel wall structure of terminal internal carotid artery (ICA) and periventricular anastomosis of MMD patients with anterior ICH.

Methods

Study design and participants

This was a cross-sectional study conducted in Xuanwu Hospital between January 2018 to January 2020. This study was approved by the Institutional Review Board of Xuanwu Hospital and written informed consent was obtained. The present study was guided by the Declaration of Helsinki.

Consecutive patients diagnosed as moyamoya disease or unilateral moyamoya disease without surgical revascularization by the guideline recommended by the Japanese research committee were enrolled. Demographic data of eligible patients including age, sex, concomitant diseases, alcohol, smoking status were recorded. Inclusion criteria containing (1) age ≥18 years; (2) subjects diagnosed as bilateral or unilateral moyamoya disease with one of the following manifestations: ischemic infarction or transient ischemic attack (ischemic type), anterior ICH with or without secondary intraventricular hemorrhage (anterior ICH type) and no symptom or subtle headache (asymptomatic type). MRI examinations were performed between one and up to three months after hemorrhage, and the delay was to ensure the adequate absorption of hematoma avoiding the severe artifact.

Exclusion criteria were as follows: (1) subjects who had undergone revascularization surgery; (2) subjects without HRMRI; (3) Subjects with cerebral ischemic/hemorrhagic region where LSAs had unclear contrast with surrounding tissue leading to failure in measurements at any slice (supplementary figure I, A); (4) subjects who had disordered LSAs generating robust networks of capillaries that made tracing incalculable (supplementary figure I, B); (5) subjects with intracranial aneurysms; (6) subjects with other diagnosed cerebral disease like brain tumor, head trauma etc. The adjudication of exclusion criteria (3) and (4) was determined by two independent assessors, and any discrepancy was solved by re-evaluation and discussion between them.

The criteria of anterior ICH were referred to as described in the JAM trial, including bleeding in putamen, caudate head, claustrum, frontal lobe, anterior half of the temporal lobe, anterior half of the periventricular area or the corpus callosum. ¹⁰

Imaging acquisition and processing

HRMRI was performed on a 3-Tesla system (Magnetom Verio; Siemens Healthineers, Erlangen, Germany), and a 32-channel head coil was used for signal reception. The technique couples a T1w 3D turbo spin-echo sequence obtained commercially, called SPACE (Sampling Perfection with Application-optimized Contrast using different flip angle Evolutions), with non-selective excitation and a trailing magnetization flip-down module. The parameters for HRMRI included: sagittal orientation, repetition time [TR]/echo time [TE] = 900/15 msec, field of view (FOV) = 170 × 170 mm², 240 slices with slice thickness of 0.53 mm, voxel size = 0.5 × 0.5 × 0.5 mm³, and scan time = 8 min. HRMRI was repeated 5 minutes after injection of contrast agent. The contrast agent was gadopentetate dimeglumine (Magnevist; Schering, Berlin, Germany), which was injected through an antecubital vein (0.1 mmol per kilogram of body weight).

Imaging analysis

Evaluation of imaging was performed by an experienced neurologist and validated by one neuroradiologist, and they were both blinded to the clinical data. Morphology of LSAs, vessel wall structure and periventricular anastomosis were evaluated on symptomatic sides of hemorrhagic/ischemic MMD patients, or unilateral hemispheres in asymptomatic patients by random selection. Twenty patients were randomly selected for analyzing the reproducibility of the quantitative measurements. Reliabilities were determined excellent when ICC > 0.75, fair to good when ICC ranges from 0.40 to 0.75 and poor when ICC < 0.40.

Morphology of LSAs

Features of LSAs including length, the number of stems and branches were measured. In addition, heubner arteries from the proximal anterior cerebral artery (ACA) were also evaluated incorporated with LSAs. Using OsiriX (University of California, Los Angeles, America), LSAs images were produced by five to six slices of minimum intensity projection (MinIP) in coronal direction with 10–15 mm thickness on pre-contrast HRMRI.

Stems longer than 5 mm and branches which were continuous, linear and clear contrast with surrounding tissue could be traced and analyzed. Stems were defined as directly originating from middle cerebral artery (MCA) or proximal ACA. When MCA or ACA was occluded, the arteries vertically sprouting from the lateral fissure where proximal MCA or ACA previously lied were identified as stems. Branches were vessels originating from stems. ¹⁸ When an artery branched within 5 mm of the origin, each stem was counted. ²² The longest branch was used when an artery branched at a distal portion. When there existed a connection between LSA and medullary artery (lenticulostriate periventricular anastomosis), the LSA would be traced from the origin to the point of anastomosis (supplementary figure II, A). Lengths of each LSA including stems and branches in one hemisphere were summed to be the total length of LSAs and were averaged as the average length of LSAs. Number of LSAs was the sum of stems and branches.

Compared with the LSA in the reference (non-MMD) hemispheres, the extremely thinner LSA was defined as shrinkage of LSAs while the prominently thicker LSA was defined as dilatation of LSA, which was exhibited in Figure 1. Therefore, the caliber of LSAs was qualitatively evaluated and categorized into three degrees: 0, shrinkage; 1, normal; 2, dilated.

Figure 1.

The degree of LSAs’ caliber. (a) shrinkage of LSA; (b) normal LSA (non-MMD hemisphere) and (c) dilatation of LSA.

Vessel wall characteristics of terminal ICA

On pre-contrast HRMRI, vessel wall measurements were performed at the terminal portion of ICA. The middle portion of cervical ICA was set as the reference normal vessel. When the terminal ICA was occluded, measurements were performed at the proximal site to the occlusion. The outer area and lumen area were measured in a plane perpendicular to the vessel length using multiplanar reformatting tool (supplementary figure II, B). Vessel wall area was defined as the difference between outer vessel area and lumen area. Remodeling index (RI) was the ratio of outer area at terminal ICA to the reference vessel. Vessel wall enhancement was qualitatively graded by comparation between pre-contrast and post-contrast T1 images, which was classified into 3 grades: grade 0, no enhancement; grade I, mild enhancement, but the signal intensity of vessel wall is less than that of the pituitary infundibulum; grade II, strong enhancement, the signal intensity of vessel wall is similar to or greater than that of the pituitary infundibulum.^23,24

Periventricular anastomosis

The evaluation of periventricular anastomosis by HRMRI was described in previous study. ²⁰ Periventricular anastomosis were classified into three types: lenticulostriate type was defined as a connection between the lenticulostriate artery and the medullary artery at the lateral corner of the frontal horn or body of the lateral ventricle; choroidal type was defined as a connection between the anterior or posterior choroidal arteries and the medullary artery beneath the lateral wall of the trigon of the lateral ventricle; thalamic type was beginning at thalamic perforating arteries and connecting to medullary or insular artery.

Statistical analysis

All statistical analyses and plots were performed using SPSS Statistics Version 23 (IBM, Armonk, New York) and R statistical software (version 4.1.2). Univariate analysis was performed to select significant demographic and imaging variates. The normality of data distribution was assessed by the Kolmogorov-Smirnov test. Continuous variables were described as mean ± SD or medians (IQRs) and categorical variables were described as proportions. To identify differences across three groups (anterior ICH, ischemic and asymptomatic groups), continuous data were analyzed by one-way ANOVA test or Kruskal-Wallis test and categorical data was analyzed by χ² test. Within three groups, Bonferroni correction was used in pairwise comparisons. Then multivariate logistic regression with forward stepwise method was applied to determine markers of anterior ICH. It adjusted for the variates with significant difference between anterior ICH and non-ICH group in univariate logistic regression. Area under curve (AUC) was calculated by receiver operator characteristic (ROC) to reflect the discriminating ability of variates selected by multivariate analysis to distinguish patients with anterior ICH from non-ICH patients.

Interrater agreements of the quantitative measurements were evaluated by intraclass correlation coefficients (ICCs). P < 0.05 and 95% confidence interval (CI) of odds ratio (OR) not containing 1 were considered to be statistically significant. All tests were 2 sided.

Data availability

The data that support the findings in this article are available from the corresponding authors on reasonable request.

Results

Inter-rater reproducibility on the quantitative measurements

For the inter-rater agreement in measurements of LSAs and vessel wall, the ICC was 0.906 (95%CI, 0.779–0.962), 0.970 (95%CI, 0.725–0.988), 0.801 (95%CI, 0.563–0.916), 0.813 (95%CI, 0.587–0.922), 0.718 (95%CI, 0.414–0.878) for total number, number of stems, number of branches, total length, average length of LSAs respectively. The ICC was 0.852 (95%CI, 0.664–0.939), 0.900 (95%CI, 0.766–0.959), 0.829 (95%CI, 0.618–0.929), 0.893 (95%CI, 0.750–0.956) for outer area, lumen area, wall area and remodeling index respectively.

Demographic characteristics

After screening 124 eligible patients, 21 patients who underwent revascularization surgery, 7 patients with cerebral aneurysms, one patient who suffered severe head trauma with cerebral bleeding, 13 patients without HRMRI examination, 2 patients with difficulty in tracing disordered LSA, and 2 patients with unclear contrast of LSAs were excluded. Finally, a total of 78 eligible MMD patients including 21 patients with anterior ICH, 31 ischemic patients and 26 asymptomatic patients were included in this study (supplementary figure III). In anterior ICH group, 15 patients suffered pure anterior ICH and 6 patients had anterior ICH with secondary intraventricular hemorrhage. The cerebral bleeding located at putamen (7/21), anterior half of periventricular area (7/21), claustrum (4/21), caudate head (2/21), anterior half of temporal lobe (1/21). Anterior ICH group had a mean age of 44.14 ± 9.01 years, and the mean age of ischemic group and asymptomatic group was 40.26 ± 10.24 years and 40.00 ± 10.80 years respectively (p > 0.05). The proportion of male patients was similar among three groups (Anterior ICH versus Ischemia versus Asymptomatic: 57.1% versus 29.0% versus 30.8%, p > 0.05).

Compared with anterior ICH group (42.9%), either ischemic group (16.1%) or asymptomatic group (19.2%) had a lower rate of smoking without significant difference (p > 0.05). No statistical difference was depicted in the prevalence of hypertension (28.6% versus 45.2% versus 42.3%, p > 0.05), diabetes (9.5% versus 12.9% versus 3.8%, p > 0.05), hyperlipidemia (28.6% versus 22.6% versus 11.5%, p > 0.05) among anterior ICH, ischemic and asymptomatic groups. The median Suzuki stage was 2 (IQR, 2–4) in anterior ICH group, 3 (IQR, 2–3) in ischemic group and 3 (IQR, 2–4) in asymptomatic group (p > 0.05). The demographic characteristics were summarized in Table 1.

Table 1.

Demographic characteristics of anterior ICH, ischemic and asymptomatic groups.

	Anterior ICH (n = 21)	Ischemic (n = 31)	Asymptomatic (n = 26)	P value
Man (%)	12/21 (57.1)	9/31 (29.0)	8/26 (30.8)	0.085
Age, mean ± SD, y/o	44.14 ± 9.01	40.26 ± 10.24	40.00 ± 10.80	0.306
Hypertension (%)	6/21 (28.6)	14/31 (45.2)	11/26 (42.3)	0.462
Diabetes (%)	2/21 (9.5)	4/31 (12.9)	1/26 (3.8)	0.489
Hyperlipidemia (%)	6/21 (28.6)	7/31 (22.6)	3/26 (11.5%)	0.333
Cigarette (%)	9/21 (42.9)	5/31 (16.1)	5/26 (19.2)	0.067
Alcohol (%)	4/21 (19.0)	2/31 (6.5)	1/26 (3.8)	0.158
Suzuki stage, median (IQR)	2 (2–4)	3 (2–3)	3 (2–4)	0.301

ICH: intracerebral hemorrhage.

Morphological features of LSAs

Compared with either ischemic group or asymptomatic group, anterior ICH group had a smaller number of LSAs (4 (IQR, 3–5) versus 6 (IQR, 4–8) versus 6.5 ((IQR, 6–8), p < 0.001), less stems (3 (IQR, 3–4) versus 5 (IQR, 4–6) versus 5 (IQR, 4–6), P < 0.001), lower mean value of total length of LSAs (90.79 ± 37.00 mm versus 138.04 ± 46.01 mm versus 170.50 ± 39.18 mm, p < 0.001). Anterior ICH group had a smaller mean value of average length compared with asymptomatic group (21.81 ± 4.33 mm versus 25.12 ± 3.13 mm, p = 0.010). Besides, the mean value of total length of LSAs in ischemic group was lower compared with asymptomatic group (p = 0.013). Number of branches (1 (IQR, 0–1) versus 1 (IQR, 0–2) versus 2 (IQR, 0–3), p > 0.05) and proportion of LSA shrinkage (57.1% versus 41.9% versus 30.8%, p > 0.05) were not significantly different across anterior ICH group, ischemic group and asymptomatic group. The features of LSAs in three groups were shown in Figure 2.

Figure 2.

Features of LSAs. LSAs features of asymptomatic group and ischemic group were shown in the a and b columns respectively; c and d columns depicted LSAs characteristics of anterior intracerebral hemorrhage. The first row showed the LSAs imaging of HRMRI; the second row represented the illustration of vascular trees constructed according to five-six slices of HRMRI imaging (MinIP); the third row displayed corresponding DSA imaging performed within one week after HRMRI examination.

HRMRI: high-resolution magnetic resonance imaging; DSA: digital subtraction angiography.

Vessel wall characteristics of terminal ICA

Larger lumen area (4.33 ± 2.02 mm² versus 2.29 ± 1.17 mm² versus 3.00 ± 1.34 mm², p < 0.001), larger outer area (16.21 ± 6.34 mm² versus 11.94 ± 3.73 mm² versus 12.80 ± 3.37 mm², p = 0.004) of terminal ICA were shown in anterior ICH group compared with either ischemic group or asymptomatic group. One-way ANOVA analysis showed statistical difference of terminal ICA wall area across three groups (11.88 ± 4.47mm² versus 9.65 ± 3.08 mm² versus 9.81 ± 2.51 mm², p = 0.046), but no discrepancy was depicted in pairwise comparison after Bonferroni correction. Remodeling index was not distinct among three groups (Anterior ICH versus Ischemia versus Asymptomatic: 0.47 ± 0.14 versus 0.42 ± 0.12 versus 0.42 ± 0.11, p = 0.846) as well as the rate of vessel wall enhancement (Anterior ICH versus Ischemia versus Asymptomatic: 33.3% versus 41.9% versus 23.1%, p = 0.323). The visualization of vessel wall characteristics of terminal ICA was shown in Figure 3.

Figure 3.

Characteristics of terminal ICA vessel wall structure. In the first column, what the red arrowhead pointed to in the white rectangle was the terminal portion of ICA, and it was also exhibited in MRA images in the third column; and the second column showed the cross-section of the terminal ICA.

Periventricular anastomosis

Lenticulostriate anastomosis was seen in 2 (9.5%) patients of anterior ICH group, 5 (16.1%) patients of ischemic group and 8 (30.0%) patients of asymptomatic group (p = 0.158). There was also no difference of choroidal anastomosis (Anterior ICH versus Ischemia versus Asymptomatic: 38.1% versus 29% versus 34.6%, p = 0.782) and thalamic anastomosis (Anterior ICH versus Ischemia versus Asymptomatic: 14.3% versus 22.6% versus 7.7%, p = 0.296) among three groups.

Features of LSAs, vessel wall characteristics and periventricular anastomosis were summarized in Table 2, and results of pairwise comparisons were listed in Table 3.

Table 2.

HRMRI features of anterior ICH, ischemic and asymptomatic groups.

Imaging features	Anterior ICH (n = 21)	Ischemic (n = 31)	Asymptomatic (n = 26)	P value
LSAs features
Number of LSAs, median (IQR)	4 (3–5)	6 (4–8)	6.5 (6–8)	<0.001*
Stems, median (IQR)	3 (3–4)	5 (4–6)	5 (4–6)	<0.001*
Branches, median (IQR)	1 (0–1)	1 (0–2)	2 (0–3)	0.128
Total length, mean ± SD, mm	90.79 ± 37.00	138.04 ± 46.01	170.50 ± 39.18	<0.001*
Average length, mean ± SD, mm	21.81 ± 4.33	22.94 ± 3.73	25.12 ± 3.13	0.010*
LSA shrinkage (%)	12/21 (57.1)	13/31 (41.9)	8/26 (30.8)	0.191
Vessel wall features of terminal ICA
Lumen area, mean ± SD, mm2	4.33 ± 2.02	2.29 ± 1.17	3.00 ± 1.34	<0.001*
Outer area, mean ± SD, mm2	16.21 ± 6.34	11.94 ± 3.73	12.80 ± 3.37	0.004*
Wall area, mean ± SD, mm2	11.88 ± 4.47	9.65 ± 3.08	9.81 ± 2.51	0.046*
Remodeling index, mean ± SD	0.47 ± 0.14	0.41 ± 0.12	0.42 ± 0.11	0.846
With enhancement (%)	7/21 (33.3)	13/31 (41.9)	6/26 (23.1)	0.323
Periventricular anastomosis
Lenticulostriate (%)	2/21 (9.5)	5/31 (16.1)	8/26(30.8)	0.158
Thalamic (%)	3/21 (14.3)	7/31 (22.6)	2/26 (7.7)	0.296
Choroidal (%)	8/21 (38.1)	9/31 (29.0)	9/26 (34.6)	0.782

*Statistical difference. ICH: intracerebral hemorrhage; LSA: lenticulostriate arteries; ICA: internal carotid artery.

Table 3.

Pairwise comparison of HRMRI imaging features.

Imaging features	Adjusted P value from pairwise comparison
	ICH vs Ischemia	ICH vs Asymptomatic	Ischemia vs Asymptomatic
LSAs features
Number of LSAs	0.003*	<0.001*	0.304
Stems	<0.001*	<0.001*	0.977
Total length	<0.001*	<0.001*	0.013*
Average length	0.847	0.010*	0.092
Vessel wall features
Lumen area	<0.001*	0.010*	0.237
Outer area	0.004*	0.034*	1.000
Wall area	0.063	0.115	1.000

*Statistical difference. ICH: intracerebral hemorrhage.

Multivariate analysis and ROC curves

To determine markers of anterior ICH, variates of which the P value was less than 0.05 in univariate analysis between ICH and non-ICH patients were selected to further perform the multivariate analysis (supplementary table I). Male, smoking, number of LSAs, total length of LSAs, average length of LSAs, lumen area, outer area and vessel wall area of terminal ICA were included in the multivariate analysis by forward stepwise method.

Multivariate analysis showed that total length of LSAs (OR 0.689, 95%CI, 0.565–0.840; p < 0.001) and lumen area of terminal ICA (OR 2.085, 95%, 1.214–3.582; p = 0.008) were associated with anterior ICH in MMD patients (Figure 4(a)).

Figure 4.

Discriminating ability of HRMRI features. (a) a forest map showing the odds ratio of potential predictors for anterior ICH analyzed by multivariate analysis; (b) ROC curves of total length of LSAs alone, terminal ICA lumen area alone and predictive probability calculated by their incorporation and (c) three-dimensional surface plot indicating the impact of LSAs length and terminal ICA lumen area on probability for anterior intracerebral hemorrhage.

ROC curves indicated that total length of LSAs had an area under curve (AUC) of 0.878 (95%CI, 0.783–0.973) and a cutoff value of 120.26 mm (sensitivity 90.50%, specificity 82.50%). The terminal lumen area had an AUC of 0.775 (95%CI, 0.656–0.895) and a cutoff value of 3.098mm² (sensitivity 81.00%, specificity 73.70%). The predictive probability calculated by their incorporation had an AUC of 0.901 (95%CI, 0.812–0.990) and a cutoff value of 0.25 (sensitivity 85.00%, specificity 82.50%). ROC curves and three-dimensional surface plot indicating the impact of LSAs length and ICA lumen area on probability for anterior intracerebral hemorrhage were shown in Figure 4(b) and (c).

Discussion

This was the first study to quantify the features of LSAs and terminal ICA in MMD patients with anterior ICH by HRMRI. Previous studies failed to reveal the mechanism of anterior ICH, our results indicated that smaller number, shorter total length of LSAs and larger luminal area of terminal ICA were in anterior ICH group compared with either ischemic or asymptomatic group. After multivariate analysis, shorter total length of LSAs and larger lumen area of ICA were independently associated with anterior ICH in MMD patients. Nevertheless, periventricular anastomosis does not seem to correlate with anterior ICH. ROC curve analysis indicated that coexistence of reduced LSAs and relatively preserved lumen area of terminal ICA could be a potential predictor for distinguishing anterior ICH risk among MMD patients.

Basal networks of MMD consist of newborn capillaries, extended perforating arteries and their anastomosis. Basal networks shift from anterior to posterior circulation with age, ¹³ and cadaveric studies indicated only a minor extent of basal networks were formed in adult MMD patients, ²⁵ which made it feasible for the measurements of perforating arteries from anterior circulation like LSAs in adult MMD patients by HRMRI.

Due to the mismatch between the sites of hemorrhage and newborn capillaries, Morioka et al. reported that newborn capillaries were not the main source of cerebral bleeding of MMD patients. ²⁶ To identify a hemorrhagic theory, increased attention was concentrated on the extended perforating arteries especially for choroidal arteries.^8,26,27 Patients with hemodynamic failure have higher risk of intracranial hemorrhage,^28,29 choroidal or thalamic arteries from posterior circulation will be extended and dilated to compensate for the ischemia of territories supplied by middle cerebral arteries, thus gradually heavier hemodynamic burden promotes the rupture of the extended arteries. However, different hemorrhagic patterns may have various pathways and risk factors. ⁸ Posterior hemorrhage mostly caused by the rupture of the dilatation and extension of choroidal or thalamic arteries,^8,10 and the formation of fragile periventricular anastomosis between extended arteries and cortical vessels lead to higher likelihood of intraventricular hemorrhage.^11,12 A recent study explored the risk factors of various hemorrhagic patterns, and found that hypertension was associated with anterior hemorrhage in patients without choroidal anastomosis. Still no distinct association between perforators and anterior hemorrhage was revealed, and the mechanisms of anterior hemorrhage need further exploration.

Our findings regarding the features of LSAs and vessel wall structure depicted by HRMRI revealed that shorter total length of LSAs and larger lumen area of terminal ICA were strongly associated with anterior ICH in MMD patients. As shown in Yamamoto’s study, although non-significantly, slightly lower proliferating degree of LSAs was demonstrated in hemorrhagic patients. ¹³ Furthermore, patients with no dilatation and no extension of LSAs were noted in the hemorrhagic group more frequently compared with non-hemorrhagic group (also non-significant). ¹¹ Despite those studies focused on both anterior and posterior intracranial hemorrhage, the results still implicated possible features of lenticulostriate arteries in hemorrhagic MMD patients which was similar to our results.

There were some plausible theories from the results of our present study. The coexistence of larger lumen area and reduced LSAs made intuitive sense like larger “pipe”, smaller “outlet” leading to more bleeding. The stenosis or degeneration of LSAs could exist in adult MMD patients, ²⁵ and previous studies regarding hypertension revealed reduced LSAs might reflect poorer function of endothelial progenitor cells that could not only promote angiogenesis but also repair the damaged vessels.^30,31 In addition, impairment in angiogenic ability was definitely found in hemorrhagic MMD patients. ³² Thus, we assumed that the reduced LSAs in anterior ICH patients may have dysfunction of endothelial progenitor cells and tend to be under a pathological and fragile status.

A large sample cohort study showed that moderate MCA stenosis was more strongly associated with increased risk of ICH due to the higher perfusion pressure compared to patients with severe stenosis at MCA. ³³ We postulated that moderate stenotic terminal ICA with insufficient collateral circulation might cause high perfusion pressure, and the reduced LSAs with limited capacity of proliferation/repairment would suffer heavier hemodynamic stress and an increased risk of bleeding. On the other hand, some anterior bleeding sites were not supplied by LSAs such as anterior half of the temporal lobe. Considering that perforating arteries from anterior and posterior circulation could be connected with each other to improve ischemia, ¹⁵ the reduction of LSAs would induce heavier hemodynamic stress of remaining vessels and lead to rupture.

Any type of periventricular anastomosis was not associated with anterior intracerebral hemorrhage in our study, which was inconsistent with previous studies.^9,12 In previous studies, approximately half or more of hemorrhagic patients suffered from intraventricular hemorrhage. ²⁰ It could be conceived that periventricular anastomosis was more likely to be associated with intraventricular hemorrhage rather than intracerebral hemorrhage, especially for the choroidal and thalamic type. Various mechanisms may exist depending on intracerebral hemorrhage or intraventricular hemorrhage of MMD patients.

The present study has some limitations. In the cohort, the reproducibility of measurements was good to excellent, while the ICC value was lower compared to the previous study. ¹⁷ In addition to the proliferation of LSAs in some MMD patients, the reasons might be the faint identification of the LSAs origins from occluded MCA and the point of anastomosis when there existed periventricular anastomosis. Patients with HRMRI exhibited disordered LSAs or severe ischemic/hemorrhagic artifact were excluded, and this might cause selection bias. The generalizability of the present results is limited due to the small sample size and the population design. This study based on Chinese adult population. First, collateral channels of hemorrhagic MMD patients may have an ethnic difference, ³⁴ whether these results could hold in MMD populations other than China is unknown. Second, application of HRMRI in pediatric MMD patients is difficult. Pediatric patients have disordered LSAs with robust networks more frequently which make the identification of LSAs difficult. ¹³ Further studies of LSAs evaluation will be implemented before ischemic or hemorrhagic events, thus the accurate predictive value of LSAs for predicting anterior ICH can be determined.

In conclusion, our results preliminarily provide new insights into the mechanism of anterior intracerebral hemorrhage. The coexistence of reduced LSAs and relatively preserved ICA lumen was associated with anterior ICH in MMD patients which could be a potential predictor of anterior hemorrhage and an indication for revascularization surgery. Expert consensus recommendations of HRMRI have been published to standardize the scanning parameters and image interpretation. ³⁵ Therefore, HRMRI can be a potential imaging technique providing adjunctive value for MMD evaluation, surveillance and treatment.