Abstract
Objective
To study the association between cerebral small vessel diseases (CSVD) and unfavorable hematoma morphology in primary intracerebral hemorrhage (ICH).
Methods
Patients with primary ICH who were admitted to West China Hospital of Sichuan University from March 2012 to January 2021 were consecutively included. The unfavorable hematoma morphology included any hypodensity, any irregularity, black hole, blend sign, Barras shape score ≥3, Barras density score ≥3, immature hematoma and combined Barras total score (CBTS) ≥4. The combined hematoma morphology score (CHMS) was evaluated by allocating 1 point for the presence of each of the mentioned unfavorable hematoma morphology. Multivariable binary logistic and ordinal regressions, together with unsupervised machine learning, were used to determine the association between CSVD and unfavorable hematoma morphology features.
Results
Univariable analysis showed that older age and hypertension were associated with white matter hyperintensities (WMH) presence. Regarding hematoma morphology, Barras density score ≥3, CBTS ≥4 and higher CHMS were associated with WMH absence (all p < 0.05). Multivariable regression indicated that lower WMH presence were significantly associated with both CBTS ≥4 and higher CHMS after correcting for confounders. Futhermore, we employed unsupervised machine learning using K-means algorithm to cluster patients into different groups according to CSVD burden, and the results showed that cluster with higher CSVD burden was less likely to be associated with unfavorable hematoma morphology such as black hole and higher CHMS after correcting for confounders.
Conclusions
Lower CSVD burden might be associated with higher incidence of unfavorable hematoma morphology features, such as CTBS ≥4 and higher CHMS.
Keywords: Cerebral small vessel disease, unfavorable hematoma morphology, primary intracerebral hemorrhage, combined hematoma morphology score, immature hematoma
Introduction
Primary intracerebral hemorrhage (ICH) accounts for approximately 10 ∼ 15% of stroke around the world, and is associated with high rates of mortality and disability [1]. Cerebral small vessel diseases (CSVD) are important causes of stroke [2]. Generally, CSVD are characterized by radiological changes, including lacunes, white matter hyperintensities (WMH), cerebral microbleeds (CMBs) and enlarged perivascular spaces (EPVS) [3]. Total CSVD score (0–4) [4] and global CSVD score (0–6) [5] proposed by Julie Staals and Marco Pasi, respectively, can reflect the effect of CSVD on the brain. Our previous study [6], together with other reports from recent literature [7], has suggested that higher CSVD score was associated with poor clinical outcome after ICH.
Meanwhile, numerous studies have suggested that unfavorable hematoma morphology is closely related to the prognosis of ICH as well. For instance, radiological signs such as black hole sign, blend sign and satellite sign can independently predict the hematoma growth in ICH [8–10], as well as mortality and morbidity [11,12]. Furthermore, hematoma maturity score, proposed by Elena Serrano [13], has larger impact on clinical outcome in comparison with the above radiological signs [13]. Specifically, the lower maturity of the hematoma, the worse the prognosis of ICH [13].
Therefore, to investigate the association between CSVD and unfavorable hematoma morphology is crucial for better understanding the mechanism of ICH. However, there is limited research on this relationship. Hence, the purpose of this research was to study the association between CSVD and unfavorable hematoma morphology in primary ICH.
Methods
Population
This study included patients with primary ICH who were admitted to West China Hospital of Sichuan University from March 2012 to January 2021. Eligible participants were included according to the following inclusion criteria: (1) aged ≥ 18 years old; (2) meeting the diagnostic criteria for primary ICH and (3) having data on both CT and MRI of the brain. Exclusion criteria were as follows: ICH secondary to craniocerebral trauma, arteriovenous malformation, intracranial aneurysm, brain tumor, hemorrhagic transformation of ischemic stroke or primary intraventricular hemorrhage. This study was approved by the Ethics Committee on Biomedical Research, West China Hospital of Sichuan University (NO.2021-803). The ethics committee waived the requirement for patient consent due to the retrospective nature of chart review.
CSVD markers and CSVD total burden assessment
We used brain MRI to assess the following CSVD markers: (1) the presence of lacunes, (2) the presence, location and number of CMBs, (3) WMH severity in deep and periventricular regions using Fazekas score [14] and (4) the severity of EPVS in basal ganglia (BG) and centrum semiovale according to STRIVE criteria [3]. We assessed CSVD score (ranging from 0 to 4) and global CSVD score (ranging from 0 to 6) to evaluate the CSVD total burden. In detail, CSVD score was assessed by awarding one point for the presence of lacunes, any CMB, BG-EPVS >10 or periventricular WMH (pWMH) extending into deep white matter (Fazekas 3) or confluent or early confluent deep WMH (dWMH) (Fazekas 2-3) as per the study by Staals et al. [4]. Global CSVD score was assessed by awarding one point for the presence of lacunes, 1–4 CMBs, BG-EPVS > 20 or moderate degree of WMH (combined WMH score 3–4), and two points for the presence of ≥5 CMBs or severe WMH (combined WMH score 5–6) [15]. Given that hypertensive arteriopathy (HA) is the major cause of primary ICH, we further utilized the HA-CSVD score [4] to additionally evaluate CSVD total burden by allocating 1 point for the presence of lacunes, ≥1 deep CMBs, >10 BG-EPVS, pWMH score 3 or dWMH score 2–3.
Unfavorable hematoma morphology assessment
In our present study, we assessed the hematoma morphology using brain CT scans. The combined hematoma morphology score (CHMS) was evaluated by allocating 1 point for the presence of each of the following features: any hypodensity, any irregularity, black hole, blend sign, Barras shape score ≥3, Barras density score ≥3, immature hematoma and combined Barras total score (CBTS) ≥4 according to the CT markers in the study by Elena Serrano et al. [13] Because of the low incidence of island sign and swirl sign, we did not incorporate them in the CHMS for this study. Theoretically, the CHMS had a range of 0–8. The definitions of the CT features were as follows: (1) any hypodensity, which was defined as an area of low density surrounded by the hyperdense hematoma, with at least an 18 Hounsfield unit (HU) difference between the two density regions; (2) any irregularity, which was defined as any irregularity at the margin, regardless of its contact with the main hematoma [13]; (3) black hole, which was defined as a relatively hypodense area encapsulated within the hyperdense hematoma, with a well-defined border and a minimum difference of 28 HU between the relatively hypodense area and the hematoma [8]; (4) blend sign, defined as a relatively hypodense area adjacent to a hyperdense region within the hematoma, with an identifiable border between the hypodense area and adjacent hyperdense region [9]; (5) Barras shape score, with a range of 1–5, was used to assess hematoma shape, where score 1 represented the most regular shape and score 5 indicated the most irregular [16]; (6) Barras density score, which was also a scale with a range of 1–5 to assess hematoma density, with score 1 representing the most homogeneous hematoma and score 5 indicating the most heterogeneous one [16]; (7) immature hematoma, which was defined as hematoma with any irregularity or hypodensity [13] and (8) CBTS, which was the sum of Barras shape score and Barras density score [13]. We additionally assessed several CT markers reported in the literature that, including (1) the number of hematoma projections or satellite foci, which was defined as a small elongated protrusion extending from the hematoma with a narrow base, or a small hematoma entirely separate from the main hematoma; (2) finger-like projections, defined as lathy protrusions arising from the hematoma where the length was greater than the width [17]; (3) round or oval hematoma; (4) satellite sign, which referred to a small hematoma with a maximal transverse diameter of less than 10 mm, completely separated from the main hematoma [10]; (5) the presence of subarachnoid hemorrhage (SAH) and (6) intraventricular hemorrhage (IVH) for each included patient. If patients had ≥1 hematomas, we evaluated the hemorrhage morphology for the largest hematoma.
ICH etiology classification
ICH etiology was determined using SMUSH-U method. In our present study, we included primary ICH by excluding patients with structural vascular malformation such as aneurysm, arteriovenous malformation, cavernous hemangioma, severe systemic disease and medication related ICH. Therefore, the etiology was classified into cerebral amyloid angiopathy (CAA)-related ICH, HA-related ICH and undetermined ICH. Specifically, patients with hemorrhage located in lobar, cortical or subcortical regions and aged ≥55 were regarded as CAA-related ICH. Those with deep or infratentorial hemorrhage and hypertension were classified as HA-related ICH per SMUSH-U method. Notably, patients with deep or infratentorial hemorrhage, with or without deep CMB, but without hypertension were also regarded as HA-related ICH according to previous study [18]. The remaining patients without a specific etiology were classified as undetermined ICH.
We further used unsupervised machine learning based on K-means algorithm to establish the association between CSVD and unfavorable hematoma morphology. Specifically, patients were grouped according to their CSVD characteristics by k-means clustering in MATLAB R2013b with 1000 replications. We entered the quantitative data for lacunes, BG-EPVS, pWMH and dWMH scores into the analysis. The number of lacunes was log-transformed by adding 1 to the lacune number. Data were evaluated using k-means clustering algorithm with k values ranging from 2 to 9. Subsequently, then total within-cluster distance was calculated for each k, and the results were plotted accordingly (Figure 1). Elbow method was used to determine the optimal number of k [19,20]. Then, we compared clinical data and imaging markers between all clusters.
Figure 1.
Total within-cluster distance and k value.
Statistics
In this study, all statistical analyses were performed using SPSS version 23 (SPSS Inc., Chicago, IL) and MATLAB R2013b software . Differences among groups were detected using Pearson chi-squared test or Fisher’s exact test for categorical variables and Student’s t test or Mann-Whitney U test for continuous variables. Binary logistic regression was used to determine the association between individual CSVD markers and the three CSVD scores with unfavorable hematoma morphology, such as black hole, blend sign, immature hematoma and CBTS ≥ 4. Ordinal regression was utilized to explore the association between CSVD and CHMS. The adjacent categoricals of CHMS were merged to simplify the ordinal regression model and satisfy the proportional odds assumption (Supplemental Table 1). For the univariable analysis of the characteristics across clusters, Bonferroni correction was applied to correct for the number of comparisons between the four clusters that were grouped based on the ‘elbow’ point of k = 4 (p = 0.05/6 = 0.008). We adjusted for age and male sex, as well as those with a p value of <0.05 in univariable analysis in multivariable regression to determine the association between CSVD and unfavorable hematoma morphology features.
Results
Overall, a total of 308 primary ICH patients were included into analysis. The flowchart was shown in Supplemental Figure 1. Among the 308 patients, there were 209 patients with HA-related ICH (67.9%), 66 patients with CAA-related ICH (21.4%), and 33 with undetermined ICH (10.7%). As shown in Table 1, there were significant differences in age (p < 0.001), hypertension (p < 0.001) and etiology distribution (p = 0.036) across the two groups by WMH absence or presence.
Table 1.
Clinical characteristics, hematoma features and CSVD between WMH absence and presence.
| Variables | Total (n = 308) | WMH absence (n = 206) | WMH presence (n = 102) | p value |
|---|---|---|---|---|
| Age, year, mean (SD) | 58.0 (14.5) | 54.3 (14.6) | 65.6 (11.1) | <0.001 |
| Male, n (%) | 218 (70.8) | 148 (71.8) | 70 (68.6) | 0.559 |
| Medical history, n (%) | ||||
| Hypertension | 218 (70.8) | 132 (64.1) | 86 (84.3) | <0.001 |
| Mellitus diabetes | 33 (10.7) | 22 (10.7) | 11 (10.8) | 0.978 |
| Hyperlipidemia | 6 (1.9) | 4 (1.9) | 2 (2.0) | 1.000 |
| Smoking | 83 (26.9) | 57 (27.7) | 26 (25.5) | 0.685 |
| Alcohol | 60 (19.5) | 43 (20.9) | 17 (16.7) | 0.380 |
| GCS on admission, median (IQR) | 15 (13–15) | 15 (14–15) | 15 (13–15) | 0.351 |
| NIHSS on admission, median (IQR) | 4 (1–8) | 4 (1–8) | 4 (2–8) | 0.212 |
| SBP, mmHg, mean ± SD* | 156.3 (27.2) | 154.5 (26.7) | 159.9 (27.8) | 0.097 |
| DBP, mmHg, mean ± SD* | 93.2 (16.6) | 93.4 (16.1) | 92.9 (17.5) | 0.799 |
| ICH etiology, n (%) | ||||
| HA | 209 (67.9) | 139 (67.5) | 70 (68.6) | 0.036 |
| CAA | 66 (21.4) | 39 (18.9) | 27 (26.5) | |
| Undetermined | 33 (10.7) | 28 (13.6) | 5 (4.9) | |
| SAH, n (%) | 19 (6.2) | 9 (4.4) | 10 (9.8) | 0.062 |
| Number of hematomas | 1 (1–1) | 1 (1–1) | 1 (1–1) | 0.413 |
| IVH, n (%) | 72 (23.4) | 44 (21.4) | 28 (27.5) | 0.234 |
| Heterogeneous hematoma intensity, n (%) | 212 (68.8) | 148 (71.8) | 64 (62.7) | 0.105 |
| Hematoma volume > 30ml, n (%)† | 49 (16.0) | 32 (15.6) | 17 (16.8) | 0.784 |
| The number of projections or satellite focuses ≥ 1, n (%)* | 128 (41.7) | 90 (43.9) | 38 (37.3) | 0.266 |
| Finger-like projections, n (%) | 9 (2.9) | 6 (2.9) | 3 (2.0) | 1.000 |
| Round or oval hematoma, n (%) | 67 (21.8) | 43 (20.9) | 24 (23.5) | 0.595 |
| Black hole, n (%) | 24 (7.8) | 19 (9.2) | 5 (4.9) | 0.183 |
| Blend sign, n (%) | 29 (9.4) | 21 (10.2) | 8 (7.8) | 0.506 |
| Immature hematoma, n (%) | 211 (68.5) | 146 (70.9) | 65 (63.7) | 0.204 |
| Satellite sign, n (%) | 55 (17.9) | 38 (18.4) | 17 (16.7) | 0.701 |
| Barras shape score ≥ 3, n (%) | 56 (18.2) | 41 (19.9) | 15 (14.7) | 0.266 |
| Barras density score ≥3, n (%) | 39 (12.7) | 33 (16.0) | 6 (5.9) | 0.012 |
| CBTS ≥ 4, n (%) | 113 (36.7) | 85 (41.3) | 28 (27.5) | 0.018 |
| CHMS, median (IQR) | 2 (0-4) | 2 (0-4) | 2 (0-4) | 0.032 |
| CSVD markers | ||||
| Lacunes ≥ 1, n (%)* | 109 (35.5) | 46 (22.3) | 63 (62.4) | <0.001 |
| CMBs ≥ 1, n (%)†† | 164 (73.5) | 86 (63.2) | 78 (89.7) | <0.001 |
| CMBs ≥ 5, n (%)†† | 86 (38.6) | 34 (25.0) | 52 (59.8) | <0.001 |
| BG-EPVS > 10, n (%)* | 160 (52.1) | 95 (46.1) | 65 (64.4) | 0.003 |
| CSVD total burden | ||||
| CSVD score (0–4), median (IQR)# | 2 (1–3) | 1 (1–2) | 3 (3–4) | <0.001 |
| Global CSVD score (0–6), median (IQR)# | 2.5 (1–4) | 1 (1–3) | 4 (3–6) | <0.001 |
| HA-CSVD score, median (IQR)# | 2 (1–3) | 1 (0–2) | 3 (2.75–4) | <0.001 |
Abbreviations: SD, standard deviation; IQR, interquartile range; GCS: Glasgow Coma Scale; NIHSS: National Institute of Health Stroke Scale; HA, hypertensive arteriopathy; CAA, cerebral amyloid angiopathy; SAH, subarachnoid hemorrhage; IVH, intraventricular hemorrhage; CBTS, Combined Barras Total Score; CHMS, the combined hematoma morphology score; CMBs, cerebral microbleeds; BG-EPVS, enlarged perivascular spaces in basal ganglia; WMH, white matter hyperintensities; CSVD, cerebral small vessel disease.
Three hundred and seven patients had data on systolic and diastolic blood pressure, the number of projections or satellite focuses, lacunes ≥ 1, and EPVS number.
Three hundred and six patients had data on hematoma volume.
Two hundred and three patients had data on CMB number.
Two hundred and two patients had data on CSVD score, Global CSVD score and HA-CSVD score.
Univariable analysis showed that older age and hypertension were associated with WMH presence (both p < 0.001). With regard to hematoma morphology, Barras density score ≥3, CBTS ≥4 and higher CHMS were associated with WMH absence (all p < 0.05). As expected, lacunes ≥1, CMBs ≥1, CMBs ≥5, BG-EPVS >10 and higher CSVD total burden were associated with WMH presence (all p < 0.05).
Multivariable analysis by adjusting for age, sex, hypertension, smoking and ICH etiology showed that older age (adjusted OR 1.024, 95% CI 1.001–1.047, p = 0.040), ICH with undetermined etiology (compared with HA-ICH, adjusted OR 3.392, 95% CI 1.456–7.902, p = 0.005) and lower incidence of WMH presence (adjusted OR 0.402, 95% CI 0.217–0.745, p = 0.004) were associated with CBTS ≥ 4. In addition, older age (adjusted OR 1.023, 95% CI 1.004–1.043, p = 0.015), ICH with undetermined etiology (adjusted OR 3.216, 95% CI 1.550–6.679, p = 0.002) and lower incidence of WMH presence (adjusted OR 0.522, 95% CI 0.314–0.865, p = 0.012) were correlated with higher CHMS (Bonferroni-corrected) (as shown in Table 2).
Table 2.
Multivariable analysis of the association between CSVD markers and unfavorable hematoma morphology.
| Variable | CBTS ≥ 4 |
CHMS* |
||
|---|---|---|---|---|
| Adusted OR (95% CI) | p value | Adusted OR (95% CI) | p value | |
| Age | 1.024 (1.001–1.047) | 0.040 | 1.023 (1.004–1.043) | 0.015 |
| Male sex | 1.090 (0.632–1.882) | 0.757 | 1.028 (0.654–1.617) | 0.904 |
| Hypertension | 1.186 (0.670–2.102) | 0.558 | 1.046 (0.649–1.685) | 0.854 |
| ICH etiology# | ||||
| CAA | 1.960 (1.041–3.693) | 0.037 | 1.844 (1.065–3.196) | 0.029 |
| Untedermined | 3.392 (1.456–7.902) | 0.005 | 3.216 (1.550–6.679) | 0.002 |
| MRI CSVD markers | ||||
| Lacune | 1.030 (0.588–1.803) | 0.918 | 0.874 (0.547–1.395) | 0.571 |
| WMH presence | 0.402 (0.217–0.745) | 0.004 | 0.522 (0.314–0.865) | 0.012 |
| BG-EPVS > 10 | 1.147 (0.670–1.964) | 0.617 | 1.087 (0.693–1.704) | 0.717 |
Abbreviations: CBTS, Combined Barras Total Score. CHMS, the combined hematoma morphology score; CAA, cerebral amyloid angiopathy; WMH, white matter hyperintensities; BG-EPVS, enlarged perivascular spaces in basal ganglia.
The adjacent categoricals of CHMS were merged to simplify the ordinal regression model and satisfy the proportional odds assumption (see supplemental table 1 for the way of merging CHMS categories).
ICH etiology was entered as categorical variables, with HA as the eferenence category.
We further used K-means algorithm to establish the association between CSVD and unfavorable hematoma morphology. Supplementary Figure 1 shows that the ‘elbow’ point of k equals 4, therefore k = 4 was chosen as the number of clusters. We compared demographics, medical history, ICH etiology, CSVD markers and unfavorable hematoma morphology between all clusters (as shown in Table 3). There were significant differences in age, hypertension, smoking, ICH etiology, lacunes ≥ 1, WMH presence, CMB presence, BG-EPVS > 10, CSVD score and black hole sign among the clusters (Bonferroni-corrected p < 0.05).
Table 3.
Characteristics across clusters (k = 4, N = 307).
| Cluster 1 (n = 81) | Cluster 2 (n = 57) | Cluster 3 (n = 79) | Cluster 4 (n = 90) | p value | |
|---|---|---|---|---|---|
| Age, year, mean (SD) | 49.1 (14.8) | 68.8 (9.6) | 55.9 (13.9) | 61.1 (12.0) | <0.001 abcde |
| Male, n (%) | 60 (74.1) | 37 (64.9) | 61 (77.2) | 59 (65.6) | 0.247 |
| Medical history, n (%) | |||||
| Hypertension | 41 (50.6) | 46 (80.7) | 56 (70.9) | 74 (82.2) | <0.001 ac |
| Diabetes mellitus | 6 (7.4) | 6 (10.5) | 11 (13.9) | 10 (11.1) | 0.618 |
| Hyperlipidemia | 2 (2.5) | 1 (1.8) | 2 (2.5) | 1 (1.1) | 0.896 |
| Smoking | 28 (34.6) | 15 (26.3) | 25 (31.6) | 15 (16.7) | 0.044 c |
| Alcohol | 22 (27.2) | 9 (15.8) | 15 (19.0) | 14 (15.6) | 0.219 |
| ICH etiology, n (%) | |||||
| HA | 48 (59.3) | 39 (68.4) | 55 (69.6) | 66 (73.3) | <0.001 abc |
| CAA | 11 (13.6) | 18 (31.6) | 18 (22.9) | 19 (21.1) | |
| Undetermined | 22 (27.2) | 0 | 6 (7.6) | 5 (5.6) | |
| Hematoma volume | 0.076 | ||||
| GCS on admission, median (IQR) | 15 (13-15) | 15 (13-15) | 15 (14-15) | 15 (13.75-15) | 0.902 |
| NIHSS on admission, median (IQR) | 4 (0–8) | 4 (2–8) | 3 (1–7) | 4 (2–9) | 0.490 |
| Lacune Lacunes ≥ 1, n (%) | 15 (18.5) | 39 (68.4) | 22 (27.8) | 33 (36.7) | <0.001 acde |
| WMH presence, n (%) | 0 | 57 (100) | 10 (12.7) | 34 (37.8) | <0.001 abcdef |
| CMB presence, n (%)†† | 24 (55.8) | 47 (90.4) | 45 (64.3) | 47 (82.5) | <0.001 acd |
| BG-EPVS > 10, n (%) | 22 (27.2) | 40 (70.2) | 40 (50.6) | 58 (64.4) | <0.001 abc |
| CSVD score (0–4), median (IQR)†† | 1 (0–2) | 4 (3–4) | 1.5 (1–2) | 2 (2–3) | <0.001 acdef |
| Black hole, n (%) | 13 (16.0) | 4 (7.0) | 4 (5.1) | 3 (3.3) | 0.012 c |
| Blend sign, n (%) | 11 (13.6) | 2 (3.5) | 7 (8.9) | 9 (10.0) | 0.258 |
| Immature hematoma | 56 (69.1) | 33 (57.9) | 62 (78.5) | 60 (66.7) | 0.079 |
| CBTS ≥ 4 | 31 (38.3) | 15 (26.3) | 37 (46.8) | 30 (33.3) | 0.084 |
| CHMS | 2 (0–4) | 2 (0–4) | 3 (1–5) | 2 (0–4) | 0.067 |
Two hundred and two patients had data on CMB presence and CSVD score (0–4).
For the comparison between the cluster 1 and cluster 2 (Bonferroni-corrected p < 0.05).
For the comparison between the cluster 1 and cluster 3 (Bonferroni-corrected p < 0.05).
For the comparison between the cluster 1 and cluster 4 (Bonferroni-corrected p < 0.05).
For the comparison between the cluster 2 and cluster 3 (Bonferroni-corrected p < 0.05).
For the comparison between the cluster 2 and cluster 4 (Bonferroni-corrected p < 0.05).
For the comparison between the cluster 3 and cluster 4 (Bonferroni-corrected p < 0.05).
Abbreviations: SD, standard deviation; IQR, interquartile range; HA, hypertensive arteriopathy; CAA, cerebral amyloid angiopathy; GCS: Glasgow Coma Scale; NIHSS: National Institute of Health Stroke Scale; WMH, white matter hyperintensities; CMBs, cerebral microbleeds; BG-EPVS, enlarged perivascular spaces in basal ganglia; CSVD, cerebral small vessel disease; CBTS, Combined Barras Total Score; CHMS, the combined hematoma morphology score.
Cluster 1 (n = 81) was the group with the youngest mean age (mean age = 49.1 years), the lowest hypertension prevalence (50.6%), the least prevalence of HA-related ICH (59.3%) and CAA-related ICH (13.6%), the lowest prevalence of lacunes ≥ 1 (18.5%), WMH presence (0%), CMB presence (55.8%) and BG-EPVS > 10 (27.2%), and the lowest CSVD score.
Cluster 2 (n = 57) was the group with the oldest mean age (mean age = 68.8 years), the highest prevalence of CAA-related ICH (31.6%), the most prevalence of lacunes ≥ 1 (68.4%), WMH presence (100%), CMB presence (90.4%) and BG-EPVS > 10 (70.2), and the highest CSVD score (median = 4).
Cluster 3 (n = 79) was the group with the second youngest age (mean age = 55.9 years), the second lowest prevalence of CMB presence (64.3%) and BG-EPVS > 10 (50.6%), and the second lowest CSVD score (median = 1.5).
Cluster 4 (n = 90) was the group with the third youngest age (mean age = 61.1 years), the most prevalence of hypertension (82.2%), the highest prevalence of HA-related ICH (73.3%) and the third least CSVD score (median = 2) (Table 3).
Table 4 shows the association between different clusters and unfavorable hematoma morphology with cluster 1 as reference. We chose cluster 1 as reference due to it was the group with the lowest CSVD burden. Overall, the cluster with lower CSVD burden was more likely to have unfavorable hematoma morphology. In detail, the incidence of black hole in cluster 4 was lower than that in cluster 1 (OR 0.180, 95% CI 0.049–0.658, Bonferroni-corrected p = 0.030), this significant difference remained by adjusting for age and sex (adjusted OR 0.138, 95% CI 0.036–0.537, Bonferroni-corrected p = 0.012), and additionally adjusting for hypertension, smoking and ICH etiology (adjusted OR 0.146, 95% CI 0.036–0.595, Bonferroni-corrected p = 0.021). Similarly, as compared to cluster 1, cluster 2 was more likely to have lower CHMS (model 1 adjusted OR 0.327, 95% CI 0.162–0.658, Bonferroni-corrected p = 0.006; model 2 adjusted OR 0.383, 95% CI 0.187–0.786, Bonferroni-corrected p = 0.027).
Table 4.
The association between different clusters and unfavorable hematoma morphology.
| Black hole |
CBTS ≥ 4 |
CHMS* |
||||
|---|---|---|---|---|---|---|
| OR (95% CI) | p | OR (95% CI) | p | OR (95% CI) | p | |
| Not adjusted | ||||||
| Cluster 2 | 0.395 (0.122–1.281) | 0.122 | 0.576 (0.275–1.208) | 0.144 | 0.557 (0.301–1.030) | 0.062 |
| Cluster 3 | 0.279 (0.087–0.897) | 0.032 | 1.421 (0.757–2.666) | 0.274 | 1.296 (0.744–2.257) | 0.361 |
| Cluster 4 | 0.180 (0.049–0.658) | 0.010 | 0.806 (0.431–1.509) | 0.501 | 0.770 (0.449–1.320) | 0.343 |
| Model 1 | ||||||
| Cluster 2 | 0.251 (0.067–0.946) | 0.041 | 0.353 (0.153–0.816) | 0.015 | 0.327 (0.162–0.658) | 0.002 |
| Cluster 3 | 0.221 (0.066–0.741) | 0.014 | 1.208 (0.631–2.312) | 0.569 | 1.062 (0.602–1.872) | 0.837 |
| Cluster 4 | 0.138 (0.036–0.537) | 0.004 | 0.600 (0.305–1.180) | 0.139 | 0.538 (0.302–0.960) | 0.036 |
| Model 2 | ||||||
| Cluster 2 | 0.256 (0.066–0.990) | 0.048 | 0.413 (0.173–0.986) | 0.047 | 0.383 (0.187–0.786) | 0.009 |
| Cluster 3 | 0.220 (0.064–0.760) | 0.017 | 1.440 (0.724–2.863) | 0.299 | 1.237 (0.689–2.223) | 0.477 |
| Cluster 4 | 0.146 (0.036–0.595) | 0.007 | 0.738 (0.359–1.519) | 0.409 | 0.658 (0.359–1.206) | 0.176 |
Ref: cluster 1.
Model 1: adjusted for age, sex.
Model 2: adjusted for Model 1 + hypertension, smoking, ICH etiology.
Abbreviations: CBTS, Combined Barras Total Score; CHMS, the combined hematoma morphology score.
The adjacent categoricals of CHMS were merged to simplify the ordinal regression model and satisfy the proportional odds assumption.
Discussion
Our study firstly explored the relationship between CSVD and unfavorable hematoma morphology, using a comprehensive set of individual hematoma morphology indicators together with the combined hematoma morphology score. The results showed that a lower CSVD burden, especialy for WMH, was more likely to have unfavorable hematoma morphologies, such as CBTS ≥4 and higher CHMS.
The relationship between CSVD and unfavorable hematoma morphology was verified by comparing clusters generated by the K-means algorithm. Specifically, we clustered the data by inputting the four common CSVD markers in ICH [3], including lacunes, BG-EPVS, pWMH and dWMH scores. The demographics, vascular risk factors and total CSVD burden across clusters were reasonable. For instance, the cluster with the lowest CSVD score had the youngest mean age, the lowest hypertension prevalence, the least prevalence of HA-related ICH and CAA-related ICH. Conversely, the cluster with the highest CSVD score had the oldest mean age and the highest prevalence of CAA-related ICH. Previous studies have suggested that age is strong risk factors for CSVD [21], and CAA generally has a high CSVD burden [2]. Therefore, our clusters generated based on the unsupervised machine learning K-means algorithm were representative of the CSVD burden.
It is noted that among all individual CSVD markers, the associations between WMH presence and unfavorable hematoma morphology were more pronounced, suggesting the involvement of WMH in the mechanism underlying unfavorable hematoma morphology. Although researches about the direct link between WMH and unfavorable hematoma morphology are lacking, previous studies have indicated that higher WMH burden was associated with a smaller hematoma volume [22]. Additionally, more severe CSVD, particularly CMBs, was associated with a lower incidence of hematoma enlargement [23]. Given that the relationships between hematoma volume and hematoma enlargement with poor hematoma morphology such as black hole sign, blend sign and island sign are substantially established [8,9,24], our results are reasonable. The mechanisms underlying the role of CSVD in unfavorable hematoma morphology might be related to the phenomena that long-term CSVD results in the thickening of small vessel walls [25,26]. Some studies have showed that the thickening small vessel walls might be related to the derangement of the blood–brain barrier, arteriolosclerosis and fibrinoid necrosis [27,28], and the resulting microvascular remodeling may function as a protective mechanism to protect the microcirculation from pulsatile barotrauma when faced with elevated systolic blood pressure and pulse pressure [29].
Consistent with previous studies [30–33], we also found that older age and hypertension might be associated with higher severity of CSVD through both traditional analysis and unsupervised machine learning. In addition, we found WMH presence might be associated with CAA-related ICH. Pantoni et al. have proposed that CAA and HA are the two common forms of CSVD in the elderly [2], and both are associated with CSVD markers [28]. It had been reported that CAA is associated with higher WMH burden compared with HA, Alzheimer’s disease and healthy older adults [34–36]. CAA is associated with amyloid β deposition within the walls of the cerebral small vessels [37]. A previous study has shown that there is a reasonably strong correlation between amyloid β severity and WMH volume [34]. Byung et al. proposed that there is a great correlation between CAA severity and reactive vascular injury [38], while decreased blood flow reactivity is associated with impaired vasodilation, resulting to higher WMH volume [39], which might explain why CAA-related ICH was associated with WMH presence.
Our study had several advantages. Firstly, our study was the first to investigate the relationship between CSVD and unfavorable hematoma morphology in primary ICH with a comprehensive set of individual hematoma morphology indicators together with the combined hematoma morphology score. Secondly, our sample size was relatively large in the study of the mechanism of CSVD in primary ICH. Thirdly, our results were verified through unsupervised machine learning by which the generated clusters were driven by the CSVD markers. More importantly, the generated clusters were representative of the CSVD burden and were reasonable.
Some limitations of our study should be considered. First, some MRI sequences were lacking in some patients, not all of the included patients had SWI sequence, resulting to the lack of data on CMB, and the three CSVD scores in some patients. Second, the patients enrolled were relatively mild because of the indications for MRI. Therefore, the generalization of our results to the entire ICH population should be cautious.
In summary, the lower burden of CSVD was associated with the higher incidence of unfavorable hematoma morphology features. Our finding is novel and needs to be verified in future studies.
Supplementary Material
Funding Statement
The study was supported by the Natural Science Foundation of Science and Technology Department of Sichuan Province (2023NSFSC1561) and Technology Innovation Research and Development Program of Chengdu Science and Technology Bureau (2024-YF05-00628-SN).
Ethics declarations
This study complies with the Declaration of Helsinki.
Ethics approval statement
The present study was approved by the Ethics Committee on Biomedical Research, West China Hospital of Sichuan University (NO.2021-803).
Patient consent statement
The ethics committee waived the requirement for patient consent due to the retrospective nature of chart review.
Disclosure statement
The authors declare no competing interests.
References
- 1.van Asch CJ, Luitse MJ, Rinkel GJ, et al. Incidence, case fatality, and functional outcome of intracerebral haemorrhage over time, according to age, sex, and ethnic origin: a systematic review and meta-analysis. Lancet Neurol. 2010;9(2):167–176. doi: 10.1016/S1474-4422(09)70340-0. [DOI] [PubMed] [Google Scholar]
- 2.Pantoni L. Cerebral small vessel disease: from pathogenesis and clinical characteristics to therapeutic challenges. Lancet Neurol. 2010;9(7):689–701. doi: 10.1016/S1474-4422(10)70104-6. [DOI] [PubMed] [Google Scholar]
- 3.Wardlaw JM, Smith EE, Biessels GJ, et al. Neuroimaging standards for research into small vessel disease and its contribution to ageing and neurodegeneration. Lancet Neurol. 2013;12(8):822–838. doi: 10.1016/S1474-4422(13)70124-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Staals J, Makin SD, Doubal FN, et al. Stroke subtype, vascular risk factors, and total MRI brain small-vessel disease burden. Neurology. 2014;83(14):1228–1234. doi: 10.1212/WNL.0000000000000837. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Pasi M, Sugita L, Xiong L, et al. Association of cerebral small vessel disease and cognitive decline after intracerebral hemorrhage. Neurology. 2021;96(2):e182–e192. doi: 10.1212/WNL.0000000000011050. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Xu M, Cheng Y, Song Q, et al. Total burden of cerebral small vessel disease in recurrent ICH versus first-ever ICH. Aging Dis. 2019;10(3):570–577. doi: 10.14336/AD.2018.0804. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Lioutas VA, Wu B, Norton C, et al. Cerebral small vessel disease burden and functional and radiographic outcomes in intracerebral hemorrhage. J Neurol. 2018;265(12):2803–2814. doi: 10.1007/s00415-018-9059-5. [DOI] [PubMed] [Google Scholar]
- 8.Li Q, Zhang G, Xiong X, et al. Black hole sign: novel imaging marker that predicts hematoma growth in patients with intracerebral hemorrhage. Stroke. 2016;47(7):1777–1781. doi: 10.1161/STROKEAHA.116.013186. [DOI] [PubMed] [Google Scholar]
- 9.Li Q, Zhang G, Huang YJ, et al. Blend sign on computed tomography: novel and reliable predictor for early hematoma growth in patients with intracerebral hemorrhage. Stroke. 2015;46(8):2119–2123. doi: 10.1161/STROKEAHA.115.009185. [DOI] [PubMed] [Google Scholar]
- 10.Shimoda Y, Ohtomo S, Arai H, et al. Satellite sign: a poor outcome predictor in intracerebral hemorrhage. Cerebrovasc Dis. 2017;44(3-4):105–112. doi: 10.1159/000477179. [DOI] [PubMed] [Google Scholar]
- 11.Zhang D, Chen J, Xue Q, et al. Heterogeneity signs on noncontrast computed tomography predict hematoma expansion after intracerebral hemorrhage: a meta-analysis. Biomed Res Int. 2018;2018:6038193. doi: 10.1155/2018/6038193. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Schreiber F, Kuschel JN, Klai M, et al. Blend sign and haemorrhage location and volume predict late recurrence and mortality in intracerebral haemorrhage patients. J Clin Med. 2023;12(19):6131. doi: 10.3390/jcm12196131. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Serrano E, López-Rueda A, Moreno J, et al. The new Hematoma Maturity Score is highly associated with poor clinical outcome in spontaneous intracerebral hemorrhage. Eur Radiol. 2022;32(1):290–299. doi: 10.1007/s00330-021-08085-4. [DOI] [PubMed] [Google Scholar]
- 14.Fazekas F, Chawluk JB, Alavi A, et al. MR signal abnormalities at 1.5 T in Alzheimer’s dementia and normal aging. AJR Am J Roentgenol. 1987;149(2):351–356. doi: 10.2214/ajr.149.2.351. [DOI] [PubMed] [Google Scholar]
- 15.Lau KK, Li L, Schulz U, et al. Total small vessel disease score and risk of recurrent stroke: validation in 2 large cohorts. Neurology. 2017;88(24):2260–2267. doi: 10.1212/WNL.0000000000004042. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Barras CD, Tress BM, Christensen S, et al. Density and shape as CT predictors of intracerebral hemorrhage growth. Stroke. 2009;40(4):1325–1331. doi: 10.1161/STROKEAHA.108.536888. [DOI] [PubMed] [Google Scholar]
- 17.Rodrigues MA, Samarasekera N, Lerpiniere C, et al. The Edinburgh CT and genetic diagnostic criteria for lobar intracerebral haemorrhage associated with cerebral amyloid angiopathy: model development and diagnostic test accuracy study. Lancet Neurol. 2018;17(3):232–240. doi: 10.1016/S1474-4422(18)30006-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Castello JP, Pasi M, Kubiszewski P, et al. Cerebral small vessel disease and depression among intracerebral hemorrhage survivors. Stroke. 2022;53(2):523–531. doi: 10.1161/STROKEAHA.121.035488. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Sammouda R, El-Zaart A.. An optimized approach for prostate image segmentation using K-means clustering algorithm with elbow method. Comput Intell Neurosci. 2021;2021(1):4553832. doi: 10.1155/2021/4553832. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Pasin O, Gonenc S.. An investigation into epidemiological situations of COVID-19 with fuzzy K-means and K-prototype clustering methods. Sci Rep. 2023;13(1):6255. doi: 10.1038/s41598-023-33214-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Thompson CS, Hakim AM.. Living beyond our physiological means: small vessel disease of the brain is an expression of a systemic failure in arteriolar function: a unifying hypothesis. Stroke. 2009;40(5):e322-330–e330. doi: 10.1161/STROKEAHA.108.542266. [DOI] [PubMed] [Google Scholar]
- 22.Wang ZJ, Zhao R, Hu X, et al. Higher cerebral small vessel disease burden in patients with small intracerebral hemorrhage. Front Neurosci. 2022;16:888198. doi: 10.3389/fnins.2022.888198. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Magid-Bernstein JR, Li Y, Cho SM, et al. Cerebral microbleeds and acute hematoma characteristics in the ATACH-2 and MISTIE III trials. Neurology. 2022;98(10):e1013–e1020. doi: 10.1212/WNL.0000000000013247. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Li Q, Liu QJ, Yang WS, et al. Island sign: an imaging predictor for early hematoma expansion and poor outcome in patients with intracerebral hemorrhage. Stroke. 2017;48(11):3019–3025. doi: 10.1161/STROKEAHA.117.017985. [DOI] [PubMed] [Google Scholar]
- 25.Ter Telgte A, van Leijsen EMC, Wiegertjes K, et al. Cerebral small vessel disease: from a focal to a global perspective. Nat Rev Neurol. 2018;14(7):387–398. doi: 10.1038/s41582-018-0014-y. [DOI] [PubMed] [Google Scholar]
- 26.Charidimou A, Pantoni L, Love S.. The concept of sporadic cerebral small vessel disease: a road map on key definitions and current concepts. Int J Stroke. 2016;11(1):6–18. doi: 10.1177/1747493015607485. [DOI] [PubMed] [Google Scholar]
- 27.Wardlaw JM. Blood-brain barrier and cerebral small vessel disease. J Neurol Sci. 2010;299(1-2):66–71. doi: 10.1016/j.jns.2010.08.042. [DOI] [PubMed] [Google Scholar]
- 28.Wardlaw JM, Smith C, Dichgans M.. Mechanisms of sporadic cerebral small vessel disease: insights from neuroimaging. Lancet Neurol. 2013;12(5):483–497. doi: 10.1016/S1474-4422(13)70060-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Mitchell GF. Effects of central arterial aging on the structure and function of the peripheral vasculature: implications for end-organ damage. J Appl Physiol (Bethesda, Md: 1985). 2008;105(5):1652–1660. doi: 10.1152/japplphysiol.90549.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Bonkhoff AK, Hong S, Bretzner M, et al. Association of stroke lesion pattern and white matter hyperintensity burden with stroke severity and outcome. Neurology. 2022;99(13):e1364–e1379. doi: 10.1212/WNL.0000000000200926. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Chen X, Chen X, Chen Y, et al. The impact of intracerebral hemorrhage on the progression of white matter hyperintensity. Front Hum Neurosci. 2018;12:471. doi: 10.3389/fnhum.2018.00471. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Giese AK, Schirmer MD, Dalca AV, et al. White matter hyperintensity burden in acute stroke patients differs by ischemic stroke subtype. Neurology. 2020;95(1):e79–e88. doi: 10.1212/WNL.0000000000009728. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Bretzner M, Bonkhoff AK, Schirmer MD, et al. MRI radiomic signature of white matter hyperintensities is associated with clinical phenotypes. Front Neurosci. 2021;15:691244. doi: 10.3389/fnins.2021.691244. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Gurol ME, Viswanathan A, Gidicsin C, et al. Cerebral amyloid angiopathy burden associated with leukoaraiosis: a positron emission tomography/magnetic resonance imaging study. Ann Neurol. 2013;73(4):529–536. doi: 10.1002/ana.23830. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Haglund M, Englund E.. Cerebral amyloid angiopathy, white matter lesions and Alzheimer encephalopathy - a histopathological assessment. Dement Geriatr Cogn Disord. 2002;14(3):161–166. doi: 10.1159/000063606. [DOI] [PubMed] [Google Scholar]
- 36.Charidimou A, Boulouis G, Haley K, et al. White matter hyperintensity patterns in cerebral amyloid angiopathy and hypertensive arteriopathy. Neurology. 2016;86(6):505–511. doi: 10.1212/WNL.0000000000002362. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Koemans EA, Chhatwal JP, van Veluw SJ, et al. Progression of cerebral amyloid angiopathy: a pathophysiological framework. Lancet Neurol. 2023;22(7):632–642. doi: 10.1016/S1474-4422(23)00114-X. [DOI] [PubMed] [Google Scholar]
- 38.Han BH, Zhou ML, Abousaleh F, et al. Cerebrovascular dysfunction in amyloid precursor protein transgenic mice: contribution of soluble and insoluble amyloid-beta peptide, partial restoration via gamma-secretase inhibition. J Neurosci. 2008;28(50):13542–13550. doi: 10.1523/JNEUROSCI.4686-08.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Peca S, McCreary CR, Donaldson E, et al. Neurovascular decoupling is associated with severity of cerebral amyloid angiopathy. Neurology. 2013;81(19):1659–1665. doi: 10.1212/01.wnl.0000435291.49598.54. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.