Island Sign on Computed Tomography Is Associated With Cerebral Amyloid Angiopathy in Patients With Intracerebral Hemorrhage

Abstract

Background

Irregular hematoma border is a prominent yet understudied imaging feature of cerebral amyloid angiopathy (CAA)‐related intracerebral hemorrhage (ICH). We investigated whether island sign and irregular shape could be an additional imaging marker of CAA‐related ICH.

Methods

This cross‐sectional analysis included a prospective cohort of consecutive patients with supratentorial ICH with both computed tomography and magnetic resonance imaging. Presence of island sign and irregular shape on computed tomography was assessed blinded to clinical and magnetic resonance imaging data. The primary outcome was probable CAA by Boston criteria v2.0. Multivariable logistic regression analysis was conducted to evaluate associations of island sign or irregular shape with probable CAA. Diagnostic accuracy of the simplified Edinburgh criteria before and after incorporating island sign or irregular shape was compared using area under the receiver operating characteristics curve and DeLong test.

Results

Among 488 patients, 51 (10.5%) patients had CAA‐ICH. Island sign (39.2% versus 4.1%, P<0.001) and irregular shape (54.9% versus 30.2%, P<0.001) were more frequent in CAA‐ICH than non‐CAA‐ICH. In multivariable analysis, island sign was independently associated with probable CAA in the overall cohort (adjusted odds ratio [OR], 4.67 [95% CI, 1.63–13.38]) and in patients with lobar ICH (OR, 4.49 [95% CI, 1.14–17.61]). Irregular shape was not related to CAA‐ICH. Among patients with lobar ICH, the simplified Edinburgh criteria incorporating island sign has improved diagnostic performance of probable CAA (area under the receiver operating characteristics curve 0.817 versus 0.754, P=0.033).

Conclusions

Island sign is associated with magnetic resonance imaging‐based diagnosis of CAA and improves the diagnostic accuracy of the simplified Edinburgh criteria, aiding in accurate identification of CAA‐related ICH.

Nonstandard Abbreviations and Acronyms

CAA

cerebral amyloid angiopathy

CMB

cerebral microbleed

cSVD

cerebral small vessel disease

FLP

finger‐like projection

ICH

intracerebral hemorrhage

WMH

white matter hyperintensity

Clinical Perspective.

What Is New?

• Irregular hematoma border has been reported as one of the most prominent imaging features of cerebral amyloid angiopathy (CAA)‐related lobar intracerebral hemorrhage. However, there is no clear definition depicting the morphological feature of CAA‐intracerebral hemorrhage.

• Island sign, characterized by multiple scattered bubble or sprout‐like small bleedings surrounding the main hematoma, is associated with probable CAA by Boston criteria v2.0 and improves the diagnostic accuracy of the simplified Edinburgh criteria based on subarachnoid hemorrhage and finger‐like projections, aiding in accurate identification of CAA‐related lobar intracerebral hemorrhage.

What Are the Clinical Implications?

• Detection and interpretation of island sign on noncontrast computed tomography may be valuable for identifying CAA‐intracerebral hemorrhage when magnetic resonance imaging is not readily available.

Approximately 80% of spontaneous intracerebral hemorrhage (ICH) cases are attributed to cerebral small vessel disease (cSVD), primarily arteriolosclerosis and cerebral amyloid angiopathy (CAA). ¹ CAA is characterized by progressive cerebrovascular deposition of amyloid β and alteration of microvascular function in the older population. ² Identification of CAA‐related ICH is critical because it contributes to an estimated 50% of all lobar ICH cases and carries highest recurrence risk among all stroke types despite risk factor control. ³

CAA can be diagnosed in vivo by its characteristic hemorrhagic and nonhemorrhagic markers on magnetic resonance imaging (MRI) according to the Boston criteria. ⁴ However, MRI is not universally accessible to all patients, particularly those with severe illnesses, contraindications, or restricted by limited scanner availability due to logistical or technical barriers. The Edinburgh criteria use APOE genotype and 2 diagnostic CT markers including subarachnoid hemorrhage (SAH) and finger‐like projections (FLPs) to stratify the risk of moderate or severe CAA pathology. These criteria showed excellent discrimination in the derivation cohort of patients with fatal ICH. ⁵ However, recent studies reported only moderate to good diagnostic accuracy and concordance for the simplified CT‐only Edinburgh criteria compared with Boston criteria in the real‐world settings. ⁶ , ⁷ , ⁸ , ⁹

Irregular hematoma border is one of the most prominent imaging features of CAA‐associated ICH, although there was absence of clear and consistent imaging marker to define this feature in previous studies. ¹⁰ Island sign, characterized by multiple scattered bubble‐ or sprout‐like small bleedings surrounding the main hematoma, is a well‐established noncontrast computed tomography (CT) marker of extreme margin irregularities predictive of hematoma expansion. ¹¹ Our previous study suggested that the prevalence of island sign was nearly twice as high in lobar ICH as in deep ICH, albeit not significant. ¹¹ Another recent study, found that island sign is significantly more frequent in lobar ICH. ¹² However, the clinical utility of island sign in differentiating ICH causes remains to be investigated.

In this study, we aimed to evaluate the associations of island sign and irregular shape with CAA‐associated ICH and to evaluate their diagnostic performance as markers of hematoma irregularity for probable CAA according to the Boston criteria v2.0. Furthermore, we sought to explore the added value of incorporating the neuroimaging markers into simplified Edinburgh criteria for the identification of probable CAA in survivors of acute ICH.

METHODS

Data Availability

Anonymized data that support the findings of this study are available by reasonable request from any qualified investigator.

Study Population

In this cross‐sectional analysis, we retrospectively analyzed data from a prospective observational cohort of consecutive patients with spontaneous ICH admitted to The Second Affiliated Hospital of Anhui Medical University (from August 2022 to March 2025) and The First Affiliated Hospital of Chongqing Medical University (from January 2016 to April 2023).

Patients with ICH were eligible for this analysis if they had diagnostic quality CT within 7 days and MRI within 90 days of symptomatic onset in line with previous studies. ⁵ , ¹³ The etiological diagnostic workup of ICH includes CT angiography, magnetic resonance angiography, or digital subtraction angiography to exclude macrovascular causes of ICH. We excluded patients with infratentorial ICH, secondary causes (eg, macrovascular causes, hemorrhagic transformation of ischemic stroke, and coagulation disorders), primary intraventricular hemorrhage, or poor imaging quality.

The study was conducted in accordance with the Declaration of Helsinki and approved by the local ethics committees. Informed consent was obtained from patients or their legal representatives. The article was prepared in adherence to the Strengthening the Reporting of Observational Studies in Epidemiology guideline. ¹⁴

Clinical Data Collection

Demographic and clinical data including age, sex, medical history, and medication use were obtained through electronic medical record review and interviews with patients or their relatives. Alcohol overuse was identified if the patients had current habitual alcohol use with >30 drinks per month or a history of binge drinking. ¹⁵

Image Acquisition and Analysis

The CT markers on admission were independently evaluated by 2 trained readers (Z. W. and X. H.) who were blinded to clinical and MRI data. Island sign and irregular shape were evaluated on standard 5 mm‐thick axial slices following the guidelines of noncontrast CT markers of ICH expansion proposed by the International NCCT [Noncontrast CT] ICH Study Group. ¹⁶ Coronal and sagittal reconstruction were not required for interpreting the 2 CT markers. Island sign was defined as the presence of either ≥3 round, oval, bubblelike scattered small bleedings surrounding but all separated from the main hematoma, or at least four small hematomas, some or all of which may connect with the main hematoma on 1 CT axial slice. ¹¹ , ¹⁶ Irregular shape was identified by the presence of ≥2 focal irregularities, either connected to or separate from the edge of main hematoma, evaluated on the axial CT slice displaying the largest area of ICH. ¹⁶ To assess the presence of SAH and FLPs, 3‐dimensional reconstruction of CT images was performed to generate coronal and sagittal planes for detailed evaluation in line with imaging criteria of Edinburgh criteria. ⁵ Discrepancies were resolved through consensus discussion and consultation with a senior neurologist (Q. L., with 18 years of experience). The interrater κ values were 0.876 for presence of island sign, 0.867 for irregular shape, 0.904 for SAH, and 0.848 for FLPs, suggesting excellent agreement.

The simplified Edinburgh criteria was applied to identify lobar ICH patients who had high probability of CAA defined by the presence of both SAH and FLPs on CT, and low probability of CAA in those with neither SAH nor FLP. ⁶ Hematoma volumes were measured with semiautomated segmentation of CT images (Analyze 12.0; https://analyzedirect.com/). The hematoma location was categorized as deep (basal ganglia or thalamus) or lobar.

Brain MRI images were acquired on 1.5 or 3.0 Tesla scanners and included at least T2‐weighted imaging, fluid‐attenuated inversion recovery, and susceptibility‐weighted imaging sequences for assessment. Evaluation of MRI‐detectable cSVD markers adhered to the Standards for Reporting Vascular Changes on Neuroimaging 2 criteria. ¹⁷ The severity of white matter hyperintensities (WMH) on axial fluid‐attenuated inversion recovery images was assessed using the Fazekas scale from 0 to 6 points, calculated by summing the deep (0 to 3) and periventricular WMH (0 to 3) scores. ¹⁸ WMH in a multispot pattern was identified if there are >10 small circular or ovoid hyperintense lesions in the bilateral subcortical white matter. ⁴ Severe perivascular spaces in the basal ganglia and centrum semiovale were defined as >20 visible perivascular spaces in either hemisphere on T2‐weighted imaging. The number and location of cerebral microbleeds (CMBs) on susceptibility‐weighted imaging were recorded according to the Microbleed Anatomical Rating Scale. ¹⁹ Cortical superficial siderosis was defined as presence of curvilinear hypointense signal patterns along the superficial layers of the cerebral cortex on susceptibility‐weighted imaging, without corresponding signal hyperintensity on FLAIR. ²⁰

The causes of cSVD‐related ICH were classified as CAA, arteriolosclerosis, mixed cSVD, or cryptogenic based on a modified version of the CLASsification system for ICH Subtypes (CLAS‐ICH) classification, which is helpful in stratifying the recurrence risk and determine comorbidity burden. ²¹ , ²² , ²³ The primary outcome was CAA‐ICH, defined as a diagnosis of probable CAA according to the Boston criteria v2.0. ⁴ Probable CAA in the Boston criteria v2.0 has been validated to diagnose CAA pathology in lobar ICH patients with an accuracy >0.90. ⁴ , ²⁴ Hemorrhagic lesions in cerebellum were not considered for any etiological category consistent with Boston criteria v2.0. ⁴ ICH attributed to arteriolosclerosis was determined in cases of deep acute ICH accompanied by at least 1 of the following cSVD features: ≥1hemorrhagic lesions (CMB or chronic ICH) confined to the deep regions (including brainstem); at least 1 lacune; moderate to severe WMH; or severe perivascular space in the basal ganglia, and in the absence of any lobar CMB or ICH. Moderate to severe WMH was defined as presence of deep Fazekas score of ≥2, or periventricular Fazekas score of 3. ²² , ²⁵ Mixed cSVD was identified by a mixture of cSVD signs in both lobar and deep regions. ICH with no MRI‐detectable cSVD markers (no moderate to severe WMH, microbleeds, lacunes, nor severe dilated perivascular space) or any other well‐defined causes were classified as cryptogenic. ¹ , ²² , ²³

Statistical Analysis

In descriptive analysis, clinical and neuroimaging characteristics were compared between patients with and without probable CAA in the overall cohort and in patients with acute lobar ICH using Pearson’s χ ² test, Fisher’s exact test, Mann–Whitney U test, or Student’s t test, as appropriate.

Multivariable binary logistic regression analyses were performed to look for independent associations of island sign or irregular shape with the diagnosis of probable CAA (versus non‐probable CAA) in the overall cohort and in the lobar ICH patients, calculating odds ratio (OR) and corresponding 95% CI. Age, sex, and biologically plausible variables with a value of P<0.1 in the univariable analysis were included in the model. Variance inflation factors (VIFs) were calculated to assess multicollinearity, and variables with a variance inflation factor >5 were removed from the multivariable models; none of the included covariates in multivariable analyses had a variance inflation factor >5.

We have conducted sensitivity analyses to test the robustness of our results. To address the potential underestimation of CAA due to the requirement of strictly lobar hemorrhages for diagnosing CAA in the Boston criteria, we expanded the definition of CAA‐related ICH to also include patients who had a high probability of CAA according to the simplified Edinburgh criteria. ⁶ , ²⁶ In addition, we reassessed the associations of island sign or irregular shape with probable CAA by further including presence of SAH and FLPs into the multivariable models.

Given that island sign may reflect multifocal vessel rupture and potentially different vessel involvement, we examined its associations with established CT and MRI markers to explore the potential pathophysiological underpinnings.

We assessed the diagnostic performance of the simplified Edinburgh criteria before and after the inclusion of island sign and irregular shape and used receiver operating characteristics curve analyses and DeLong test to compare the overall discrimination ability for probable CAA.

The statistical significance was set at a P value of 0.05. Data management and statistical analysis were conducted using SPSS (version 25.0; www.spss.com) and MedCalc (version 20.022; www.medcalc.org).

RESULTS

The final study population comprised 488 patients (360 with deep ICH and 128 with lobar ICH) categorized as follows: 51 (10.5%) with CAA‐ICH, 133 (27.3%) with arteriolosclerosis, 258 (52.9%) with mixed cSVD, and 46 (9.4%) with cryptogenic ICH (Figure 1). MRI scans on 3.0 Tesla scanners were completed on 85.7% of the study participants.

Figure 1. Study flow chart.

CAA indicates cerebral amyloid angiopathy; cSVD, cerebral small vessel disease; ICH, intracerebral hemorrhage; and IVH, intraventricular hemorrhage.

Baseline clinical and imaging characteristics of patients with and without a diagnosis of probable CAA in the overall cohort are compared in Table 1. Patients with CAA‐ICH according to the Boston criteria v2.0 were older (median age 72 [interquartile range, 65–78] years versus 61 [interquartile range, 52–71] years, P<0.001), less likely to have history of hypertension (52.9% versus 89.2%, P<0.001) and smoking (13.7% versus 32.5%, P=0.006), had longer time from onset to CT scan (median 0.9 [interquartile range, 0.4–2.0] days versus 0.4 [interquartile range, 0.1–1.0] days, P<0.001), and exhibited larger hematoma volumes (median 25.0 [interquartile range, 12.4–41.1] mL versus 7.8 [interquartile range, 3.1–14.9] mL, P<0.001). The prevalence of island sign (39.2% versus 4.1%, P<0.001) and irregular shape (54.9% versus 30.2%, P<0.001) were significantly more common in the group with probable CAA group. Representative images of hematoma morphological features in CAA‐ICH and other causal subtypes were displayed in Figure 2. The frequencies of island sign and irregular shape stratified by locations and causal subtypes were detailed in Figure S1.

Table 1.

Comparison of Baseline Characteristics Between Patients With Probable and Non‐Probable CAA in the Overall Population

	Total (n=488)	Non‐probable CAA (n=437)	Probable CAA (n=51)	P value
Clinical characteristics
Age, y, median (IQR)	62 (53–72)	61 (52–71)	72 (65–78)	<0.001
Male sex, n (%)	326 (66.8)	296 (67.7)	30 (58.8)	0.201
Hypertension, n (%)	417 (85.5)	390 (89.2)	27 (52.9)	<0.001
Diabetes, n (%)	110 (22.5)	100 (22.9)	10 (19.6)	0.596
Hypercholesterolemia, n (%)	165 (33.8)	152 (34.8)	13 (25.5)	0.184
Current smoking, n (%)	149 (30.5)	142 (32.5)	7 (13.7)	0.006
Alcohol overuse, n (%)	71 (14.5)	65 (14.9)	6 (11.8)	0.551
Previous ischemic stroke, n (%)	79 (16.2)	72 (16.5)	7 (13.7)	0.614
Previous ICH, n (%)	46 (9.4)	43 (9.8)	3 (5.9)	0.456
Previous antiplatelet, n (%)	56 (11.5)	51 (11.7)	5 (9.8)	0.692
Previous anticoagulant, n (%)	6 (1.2)	5 (1.1)	1 (2.0)	0.486
Imaging characteristics
Time from onset to computed tomography, d, median (IQR)	0.4 (0.1–1.0)	0.4 (0.1–1.0)	0.9 (0.4–2.0)	<0.001
Time from onset to magnetic resonance imaging, d, median (IQR)	2.6 (0.3–6.7)	2.4 (0.3–6.7)	4.4 (0.8–6.7)	0.122
Lobar ICH, n (%)	128 (26.2)	77 (17.6)	51 (100.0)	<0.001
Baseline ICH volume, mL, median (IQR)	8.5 (3.3–17.2)	7.8 (3.1–14.9)	25.0 (12.4–41.1)	<0.001
Presence of intraventricular hemorrhage, n (%)	132 (27.0)	117 (26.8)	15 (29.4)	0.688
Presence of subarachnoid hemorrhage, n (%)	52 (10.7)	20 (4.6)	32 (62.7)	<0.001
Finger‐like projection, n (%)	67 (13.7)	39 (8.9)	28 (54.9)	<0.001
Irregular shape, n (%)	160 (32.8)	132 (30.2)	28 (54.9)	<0.001
Island sign, n (%)	38 (7.8)	18 (4.1)	20 (39.2)	<0.001

CAA indicates cerebral amyloid angiopathy; ICH, intracerebral hemorrhage; and IQR, interquartile range.

Figure 2. Representative images of CT morphological markers in ICH with different underlying small vessel disease types.

A and D, Patient with CAA‐ICH with right lobar ICH, island sign (arrowheads) on computed tomography, and multiple strictly lobar CMBs. B and E, Patient with ICH attributed to arteriolosclerosis with left putaminal regular‐shaped hematoma and strictly deep CMBs. C and F, Patient with mixed cSVD with left regular‐shaped lobar ICH and concurrent deep CMBs. CAA indicates cerebral amyloid angiopathy; CMB, cerebral microbleed; cSVD, cerebral small vessel disease; CT, computed tomography; and ICH, intracerebral hemorrhage.

Among 128 patients with lobar ICH, 51 (39.8%) were classified as having probable CAA, 72 (56.3%) as mixed cSVD, and 5 (3.9%) as cryptogenic. The time from onset to CT scan between patients with and without CAA was similar in patients with lobar ICH. Patients with lobar ICH and probable CAA had significantly higher proportion of island sign (39.2% versus 7.8%, P<0.001) and numerically higher frequency of irregular shape (54.9% versus 40.3%, P=0.104) than those with lobar ICH but did not meet the criteria for probable CAA (Table 2).

Table 2.

Comparison of Baseline Characteristics Stratified by Probable Versus Non‐Probable CAA Among Patients With Lobar ICH

	Total (n=128)	Non‐probable CAA (n=77)	Probable CAA (n=51)	P value
Clinical characteristics
Age, y, median (IQR)	70 (57–77)	69 (55.5–75.5)	72 (65–78)	0.013
Male sex, n (%)	85 (66.4)	55 (71.4)	30 (58.8)	0.139
Hypertension, n (%)	97 (75.8)	70 (90.9)	27 (52.9)	<0.001
Diabetes, n (%)	28 (21.9)	18 (23.4)	10 (19.6)	0.614
Hypercholesterolemia, n (%)	36 (28.1)	23 (29.9)	13 (25.5)	0.589
Current smoking, n (%)	35 (27.3)	28 (36.4)	7 (13.7)	0.005
Alcohol overuse, n (%)	19 (14.8)	13 (16.9)	6 (11.8)	0.425
Previous ischemic stroke, n (%)	24 (18.8)	17 (22.1)	7 (13.7)	0.236
Previous intracerebral hemorrhage, n (%)	15 (11.7)	12 (15.6)	3 (5.9)	0.095
Previous antiplatelet, n (%)	16 (12.5)	11 (14.3)	5 (9.8)	0.453
Previous anticoagulant, n (%)	4 (3.1)	3 (3.9)	1 (2.0)	1.000
Imaging characteristics
Time from onset to computed tomography, d, median (IQR)	0.8 (0.3–1.8)	0.8 (0.2–1.8)	0.9 (0.4–2.0)	0.561
Time from onset to magnetic resonance imaging, d, median (IQR)	4.4 (0.9–7.3)	4.4 (0.9–7.6)	4.4 (0.8–6.7)	0.503
Baseline ICH volume, mL, median (IQR)	17.7 (7.6–26.9)	13.9 (3.5–22.0)	25.0 (12.4–41.1)	<0.001
Presence of intraventricular hemorrhage, n (%)	29 (22.7)	14 (18.2)	15 (29.4)	0.137
Presence of subarachnoid hemorrhage, n (%)	41 (32.0)	9 (11.7)	32 (62.7)	<0.001
Finger‐like projection, n (%)	50 (39.1)	22 (28.6)	28 (54.9)	0.003
Irregular shape, n (%)	59 (46.1)	31 (40.3)	28 (54.9)	0.104
Island sign, n (%)	26 (20.3)	6 (7.8)	20 (39.2)	<0.001

CAA indicates cerebral amyloid angiopathy; ICH, intracerebral hemorrhage; and IQR, interquartile range.

In multivariable analysis, the presence of island sign (OR, 4.67 [95% CI, 1.63–13.38], P=0.004) was independently associated with probable CAA in the overall population after adjusting for age, sex, hypertension, current smoking, time from onset to CT scan, and hematoma volume. In contrast, there was no correlation between irregular shape and CAA‐ICH (OR, 0.78 [95% CI, 0.33–1.84], P=0.577). When the analysis was restricted in patients with lobar ICH, island sign remained associated with probable CAA (OR, 4.49 [95% CI, 1.14–17.61], P=0.032) after controlling for age, sex, hypertension, current smoking, previous ICH, and hematoma volume (Table 3). We repeated the multivariable analysis after including 6 additional patients who were not diagnosed with probable CAA but a high probability of CAA based on the simplified Edinburgh criteria accounting for a broader definition of CAA. Island sign was consistently associated with diagnosis of CAA in the overall cohort (OR, 5.93 [95% CI, 2.12–16.57], P=0.001) and in the lobar ICH patients (OR, 8.71 [95% CI, 1.84–41.30], P=0.006) after adjusting for the same covariates, whereas irregular shape was not associated with CAA‐related ICH (Table S1). After further adjusting for presence of SAH and FLPs, island sign, but not irregular shape, remained associated with probable CAA (Table S2).

Table 3.

Univariable and Multivariable Binary Logistic Regression Analyses of the Associations of Island Sign or Irregular Shape With the Diagnosis of Probable CAA Versus Non‐Probable CAA

	Crude OR (95% CI)	P value	Adjusted OR (95% CI)	P value
All ICH*
Island sign	15.02 (7.21–31.28)	<0.001	4.67 (1.63–13.38)	0.004
Irregular shape	2.81 (1.56–5.07)	0.001	0.77 (0.33–1.80)	0.550
Lobar ICH†
Island sign	7.63 (2.79–20.86)	<0.001	4.49 (1.14–17.61)	0.032
Irregular shape	1.81 (0.88–3.69)	0.105	0.49 (0.16–1.50)	0.214

CAA indicates cerebral amyloid angiopathy; ICH, intracerebral hemorrhage; and OR, odds ratio.

^*Adjusted for age, sex, hypertension, current smoking, time from onset to computed tomography, and baseline ICH volume.

^† Adjusted for age, sex, hypertension, current smoking, previous ICH, and baseline ICH volume.

Associations between island sign and imaging characteristics on CT and MRI among patients with lobar ICH are reported in Table 4. Among patients with lobar ICH, island sign was associated with larger hematoma volume and higher frequencies of SAH, FLPs, and strictly lobar CMBs, with a trend toward more cortical superficial siderosis.

Table 4.

Association Between Island Sign and Imaging Characteristics in Lobar ICH

	Without island sign (n=102)	With island sign (n=26)	P value
Time from onset to computed tomography, d, median (IQR)	0.9 (0.3–2.0)	0.8 (0.2–1.5)	0.337
Hematoma volume, mL, median (IQR)	13.9 (3.6–22.6)	39.3 (26.2–65.4)	<0.001
Presence of subarachnoid hemorrhage, n (%)	24 (23.5)	17 (65.4)	<0.001
Finger‐like projection, n (%)	27 (26.5)	23 (88.5)	<0.001
WMH Fazekas score, median (IQR)	4 (2–6)	3 (2–4)	0.395
Multispot WMH, n (%)	28 (27.5)	9 (34.6)	0.472
Severe centrum semiovale perivascular space, n (%)	24 (23.5)	8 (30.8)	0.447
Presence of CMBs, n (%)	87 (85.3)	22 (84.6)	0.931
Strictly lobar CMBs, n (%)	23 (22.5)	15 (57.7)	<0.001
Lobar CMB count, median (IQR)	3 (0–12)	2 (1–4)	0.312
Presence of cortical superficial siderosis, n (%)	24 (23.5)	11 (42.3)	0.055

Categorical variables were presented as number (percentage) and compared using Pearson χ ² or Fisher exact test as appropriate. Continuous variables were presented as median (interquartile range) and compared using Mann–Whitney U test. CMB indicates cerebral microbleed; ICH, intracerebral hemorrhage; IQR, interquartile range; and WMH, white matter hyperintensity.

For patients with lobar ICH, island sign yielded a sensitivity of 0.39 (95% CI, 0.26–0.54) and a specificity of 0.92 (95% CI, 0.83–0.97) for diagnosing probable CAA. Irregular shape had a sensitivity of 0.55 (95% CI, 0.40–0.69) and a specificity of 0.60 (95% CI 0.48–0.71) for diagnosing probable CAA. The overall discrimination ability for probable CAA was higher when incorporating island sign (AUC, 0.817 [95% CI, 0.739–0.896] versus AUC, 0.754 [95% CI, 0.662–0.846], P=0.033) rather than irregular shape (P=0.639) into the simplified Edinburgh criteria (Figure 3 and Table S3). By combining island sign with SAH and FLPs, 6 additional patients were classified as having a high probability of CAA. For rule‐in category, the high‐risk category defined by the presence of at least any 2 of CT markers including SAH, FLPs, and island sign led to a 6% increase in sensitivity (0.53 [95% CI, 0.39–0.67] versus 0.47 [95% CI, 0.33–0.61]), at the expense of a 4% drop in specificity (0.88 [95% CI, 0.78–0.94] versus 0.92 [95% CI, 0.83–0.97]). The low‐risk category defined as absence of all 3 CT markers had a higher sensitivity (0.76 [95% CI 0.62–0.87] versus 0.71 [95% CI, 0.56–0.82]) and similar specificity of (0.67 [95% CI, 0.56–0.78] versus 0.68 [95% CI, 0.56–0.78]) to rule out probable CAA.

Figure 3. Receiver operating characteristic curve for the simplified Edinburgh criteria before and after inclusion of island sign or irregular shape.

The area under the curve for the simplified Edinburgh criteria in predicting probable cerebral amyloid angiopathy before and after incorporating island sign or irregular shape. AUC indicates area under the curve.

DISCUSSION

Our analysis of a large cohort of patients with spontaneous ICH demonstrated that island sign and irregular shape detected by noncontrast CT is significantly more common in CAA‐ICH compared with non‐CAA ICH. In multivariable analysis, island sign, but not irregular shape, was associated with CAA‐ICH. Island sign in lobar ICH was related to SAH, FLPs, and strictly lobar CMBs. Integrating island sign with the simplified Edinburgh criteria improved the discriminative performance for probable CAA in patients with lobar ICH by gaining increased sensitivity with only a slight reduction in specificity. Given that CT is the most widely used diagnostic imaging modality for ICH, detection and interpretation of island sign on CT may be valuable for distinguishing CAA‐ICH.

We found that the prevalence of island sign and irregular shape was substantially higher in CAA‐related ICH compared with non‐CAA ICH, including those attributed to arteriosclerosis, mixed microangiopathy, or cryptogenic causes. These results are consistent with recent studies that highlighted irregularity as a neuroimaging marker suggestive of underlying CAA. A previous meta‐analysis found that an irregular border of hematoma is one of the most prominent neuroimaging characteristics of CAA‐related lobar ICH, whereas currently no well‐defined imaging marker for this feature has been established in the field of CAA diagnosis. ¹⁰ Another recent study reported a higher prevalence of island sign in lobar ICH than nonlobar ICH (26.2% versus 9.2%), supporting our findings of high occurrence of island sign in CAA‐related lobar ICH. ¹² However, the underlying cause could not be determined due to absence of MRI data. In contrast, the definition of irregular shape, which requires the presence of only >2 morphological variations, is less stringent than that of island sign. Consequently, irregular shape seems to be less specific than island sign for differentiating CAA from other cSVD types, despite higher prevalence in CAA‐ICH in the univariable analysis.

There is little knowledge about the pathophysiology of island sign in CAA‐ICH. Hypothetically, one could presume that the lobar location may predispose to larger ICH, and higher volume tolerance of lobar ICH may also delay the time from symptom onset to ICH diagnosis, ²⁷ , ²⁸ which could lead to the misattribution of island sign. After adjusting for multiple covariates including hematoma volume and time from onset to imaging scan, our results demonstrated that island sign remained associated with CAA‐ICH, whereas no significant correlation was found between irregular shape and probable CAA. This finding suggests that the presence of extreme shape variations as represented by island sign may not merely indicate larger or expanding hematomas but could also serve as a specific neuroimaging marker of CAA pathology.

Our analyses of CT and MRI markers showed strong associations of island sign with SAH, FLPs, and strictly lobar CMBs. In addition, a numerically higher prevalence of cortical superficial siderosis, a hemorrhagic‐prone marker suggestive of repeated bleeding in the subarachnoid space in CAA, was observed, though this association did not reach statistical significance, possibly due to the limited sample size of patients with CAA in our cohort. Previous studies have found that FLP is associated with presence of strictly lobar CMBs and APOE ε4 allele and may reflect predominant parenchymal vessel involvement in CAA pathology, whereas SAH is more frequent in patients with cortical superficial siderosis and APOE ε2 allele, which are markers suggestive of leptomeningeal small‐vessel pathology. ⁷ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ The observed associations of island sign with multiple CAA imaging markers suggest that island sign may represent a more advanced stage of CAA‐related small vessel pathology rather than a feature confined to severe parenchymal involvement. A recent study found that APOE ε4 possession, which contributes to parenchymal vascular amyloid accumulation, is associated with island sign in lobar but not in deep ICH. ¹² This supports the hypothesis that the mass effect of primary hemorrhage is more likely to trigger secondary ruptures of adjacent fragile small vessels and lead to multifocal bleeding according to the avalanche model on the pathological basis of extensive vascular wall fragmentation and loss of smooth muscle cells due to amyloid deposition. ³ The concurrence of island sign and SAH may also indicate both leptomeningeal and parenchymal small vessel ruptures, predisposing to a particularly high risk of hematoma expansion. ²⁹ , ³⁴ , ³⁵ Future studies are needed to elucidate the neuropathologic correlates of island sign in patients with CAA.

Differentiation of CAA from non‐CAA related ICH carries significant clinical implications because CAA is associated with highest risk of recurrence and dementia among all ICH causes. ³ , ³⁶ CAA‐related ICH may also herald different dynamics of hematoma expansion with prolonged time window compared with other ICH subtypes. ³⁷ MRI‐based Boston criteria that rely on hemorrhagic and non‐hemorrhagic cSVD markers are the current standard for in vivo diagnosis of CAA. Probable CAA, as defined by the Boston criteria v2.0, demonstrates an accuracy exceeding 0.90 for confirming CAA pathology. ⁴ , ²⁴ However, MRI may be contraindicated or poorly tolerated in patients with critical illness. MRI including blood‐sensitive sequences remains underused in resource‐limited settings. Emerging data showed that the simplified CT‐only Edinburgh criteria, when applied in a real‐world setting, achieved an estimated overall discriminatory performance of probable CAA ranging from 0.62 to 0.76, highlighting the need for further investigations into the imaging characteristics of CAA‐ICH to increase the ability of diagnosis of CAA based on CT. ⁶ , ⁷ , ⁸ Our data showed that inclusion of island sign into simplified Edinburgh criteria could increase the sensitivity for diagnosis of probable CAA. However, as SAH, FLPs, and island sign are less frequently observed in small hematomas, their absence is likely less effective to rule out CAA with small hematoma. This finding underscores the importance of MRI in survivors of ICH with suspected CAA. Beyond improving diagnosis, CT markers of CAA may also provide additional prognostic information. Island sign is associated with hematoma expansion and poor clinical outcomes. ¹¹ Recent studies found that SAH is associated with hematoma expansion ³⁴ and increased risk of recurrent ICH. ³² , ³³ It is of potential interest to investigate the role of island sign as a prognostic marker for patients with acute lobar ICH.

The main strength of our study lies in the detection of cSVD markers to classify ICH causes in a large cohort of patients with ICH with complete multimodal CT and MRI data. However, several limitations should be considered when interpreting our findings. First, we were unable to analyze the histopathologic correlations between island sign and CAA. The probable CAA category as defined by the Boston criteria v2.0 demonstrates high specificity for CAA pathology in ICH cases, while the sensitivity of MRI‐based diagnosis remains suboptimal. A subset of mixed and cryptogenic cases may still harbor underlying CAA pathology, underscoring the need for pathological validation. ²⁴ Second, our study population was restricted to participants who could tolerate MRI scan. Patients with island sign are more likely to have larger hematomas, which may result in more severe clinical symptoms and consequently make them less likely to undergo MRI scan, which may induce selection bias. Therefore, the prevalence of island sign may be underestimated. Third, this study was conducted in Chinese population, and our results may not be generalizable to other racial or ethnic groups. Although a recent meta‐analysis reported a similar prevalence of CAA in Asian and Western populations, Asian populations more frequently exhibited deep or mixed CMB distribution rather than strictly lobar CMBs, which may affect the diagnosis of CAA according to the Boston criteria. ³⁸ Finally, residual confounding factors cannot be fully excluded due to the inherent limitations of an observational study design.

CONCLUSIONS

Our study found that island sign is associated with CAA‐related ICH and may serve as a CT marker supporting more accurate diagnosis of CAA among patients with acute lobar ICH, particularly when MRI is unavailable. Further studies incorporating histopathological validation and longitudinal follow‐up are needed to establish the diagnostic and prognostic value of the island sign in CAA‐ICH.

Sources of Funding

This study was supported by National Natural Science Foundation of China (No. 82471368), Excellent Research and Innovation Team Project of Anhui Province (No. 2024AH010014), Clinical and Translational Research Project of Anhui Province (No. 202427b10020090, No. 202427b10020053), Health Research Program of Anhui (No. 2024Aa40015), and Research Fund of Anhui Institute of Translational Medicine (No. 2022zhyx‐C38).

Disclosures

The authors declare no conflict of interest.

Supporting information

Tables S1–S3

Figure S1