Automated quantification of reduced sulcal volume identifies early brain injury after aneurysmal subarachnoid hemorrhage

Abstract

Background and Purpose:

Early brain injury (EBI) may be a more significant contributor to poor outcome after aneurysmal subarachnoid hemorrhage (aSAH) than vasospasm and delayed cerebral ischemia. However, studying this process has been hampered by lack of a means of quantifying the spectrum of injury. Global cerebral edema (GCE) is the most widely accepted manifestation of EBI, but is currently assessed only through subjective, qualitative or semi-quantitative means. Selective sulcal volume (SSV), the CSF volume above the lateral ventricles, has been proposed as a quantitative biomarker of GCE, but is time-consuming to measure manually. Here we implement an automated algorithm to extract SSV and evaluate the age-dependent relationship of reduced SSV on early outcomes after aSAH.

Methods:

We selected all adults with aSAH admitted to a single institution with imaging within 72 hours of ictus. Scans were assessed for qualitative presence of GCE. SSV was automatically segmented from serial CTs using a deep learning -based approach. Early SSV was the lowest SSV from all early scans. Modified Rankin Scale (mRS) of 4–6 at hospital discharge was classified as a poor outcome.

Results:

244 aSAH patients were included. 65 (27%) had GCE on admission while 24 developed it subsequently within 72 hours. Median SSV on admission was 10.7-ml but frequently decreased, with minimum early SSV being 3.0-ml (IQR 0.3–11.9). Early SSV below 5-ml was highly predictive of qualitative GCE (AUC 0.90). Reduced early SSV was an independent predictor of poor outcome, with a stronger effect in younger patients.

Conclusions:

Automated assessment of SSV provides an objective biomarker of GCE that can be leveraged to quantify EBI and dissect its impact on outcomes after aSAH. Such quantitative analysis suggests that GCE may be more impactful to younger SAH patients.

Introduction

Aneurysmal subarachnoid hemorrhage (aSAH) comprises only five percent of all strokes but afflicts younger people and carries a particularly high mortality.^{1, 2} Moreover, many SAH survivors suffer long-term cognitive deficits that negatively impact their quality of life and return to independence.^{3, 4} For decades, much of this morbidity and mortality was attributed to the consequences of vasospasm and delayed cerebral ischemia (DCI). However, interventions that effectively treated vasospasm did not improve functional outcomes.⁵ In recent years, increasing evidence has suggested that early brain injury (EBI), a result of pathological processes triggered by aneurysmal rupture and developing within 72 hours of ictus, also have a major contribution to impaired recovery following aSAH.⁶ However, a complete understanding of the extent and impact of EBI remains insufficient due to the lack of an easily accessible and reliable biomarker that captures the entire spectrum of severity and its impact on clinical course after SAH.

Research into DCI has been advanced by the ability to measure both vasospasm and the consequences of ischemia, primarily cerebral infarction. At the present, no consensus on how to define and measure EBI exists. The span of literature defines EBI using either: 1) assessment of the severity of aneurysmal rupture, with clinical grading scales such as the Hunt and Hess or World Federation of Neurosurgical Societies’ (WFNS), or occurrence of ictal loss of consciousness; 2) neuroimaging detection of global cerebral edema (GCE); 3) invasive neuromonitoring for brain tissue hypoxia or metabolic distress; or 4) worsening neurological exam during the first few days after admission.^7–11

The most widely accepted and commonly utilized biomarker for EBI is the presence of GCE on imaging obtained either on or shortly after admission. GCE is felt to represent the downstream consequences of EBI, as the cascade of SAH-induced injury mechanisms results in cellular injury, neuroinflammation, blood-brain barrier permeability, and water accumulation.⁶ GCE was initially defined by Claassen et al., through a dichotomous assessment focusing on sulcal effacement (qualitative GCE).⁹ Assessment of sulcal effacement was recently refined by Ahn et al., through a semi-quantitative scale called the Subarachnoid Hemorrhage Early Brain Edema Score (SEBES).⁸ The incidence of GCE reported using these two methods ranges from 6–57%.^{8, 9, 12, 13} This broad variability may be attributed to the subjective nature of both of these assessment tools. Furthermore, these methods are limiting in only capturing the most severe cases of edema, rather than quantifying the full spectrum of EBI. As brain injury after SAH likely affects all patients to varying degrees, modulated by biologic and clinical factors, a means of fully quantifying GCE would greatly facilitate its study.

Recently, Choi and colleagues described a new method for assessing GCE after SAH. This approach manually outlined the CSF in all sulci above the level of the ventricle, obtaining Selective Sulcal Volume (SSV). Their preliminarily data demonstrated that SSV was lower in those with qualitative GCE and that reduced SSV was associated with poor outcome in a small cohort of SAH patients.¹⁴ However, manually quantifying sulcal volume is labor-intensive and time-consuming, which limits the feasibility of its application to the study of EBI. Moreover, that study and those utilizing SEBES did not account for age as a confounding factor; younger SAH patients are likely to have lower sulcal volumes regardless of their actual degree of EBI. In addition, this study, as well as others, focused on assessing GCE at the time of admission, neglecting the dynamic and evolving nature of EBI in the days following ictus.

In the present study, we aimed to expand upon the initial findings of Choi and colleagues in four fundamental ways: 1) Fully automating the process of quantifying SSV in SAH patients to facilitate its assessment; 2) Examining reduction in SSV over several days after SAH, not only on admission; 3) Accounting for the impact of age on SSV measurements; and 4) Validating the association between reduced SSV and outcome in a larger, independent cohort, evaluating and adjusting for the effect of patient age.

Methods

Subject Selection

The data that support the findings of this study are available from the corresponding author upon reasonable request. De-identified data on all patients admitted to Barnes-Jewish Hospital with nontraumatic subarachnoid hemorrhage from January 2014 to March 2018 were extracted from an institutional SAH registry. Inclusion criteria for this study were as follows: 1) adult patients aged 18 years or older; 2) aSAH confirmed by digital subtraction angiography or CT angiography or presumed aSAH (based on pattern of bleeding on CT) who did not undergo confirmatory angiography due to poor clinical status. Patients were excluded under the following conditions: 1) initial CT obtained greater than 72 hours after ictus; 2) non-aneurysmal SAH (e.g. trauma, arteriovenous malformation rupture) or angiography-negative SAH. This retrospective study was approved by the Institutional Review Board with waiver of informed consent.

Clinical and Radiographic Variables

Baseline demographic data obtained included age, gender and race. Clinical features upon presentation were recorded by a trained research nurse, including Hunt and Hess, WFNS scores, and loss of consciousness at ictus. Episodes of worsening neurological exam and presence of any confounding causes were prospectively recorded and reviewed by an attending neurointensivist. Hydrocephalus was defined as ventricular enlargement with associated symptoms, leading to placement of an external ventricular drain (EVD). Osmotic therapies were utilized in those with poor or worsening neurological status in the presence of GCE or raised intracranial pressure (ICP) and were not administered prophylactically. A diagnosis of DCI was made according to previously established criteria.¹⁵ Outcome was determined using the modified Rankin Scale (mRS) at time of hospital discharge, in which a score of 4–6 was considered poor outcome.⁸

Imaging Assessment of Early Brain Injury

All CTs were transferred from the radiology server to a research repository (SNIPR, the Stroke Neuroimaging Phenotype Repository) for quantitative analysis. Each CT was first manually reviewed by one of three study investigators and excluded for significant motion artifact. If several scans were available at a single time point, non-contrast soft-tissue windowed CTs with slice thickness between 3 and 5-mm were selected for consistency of image analysis. Each eligible CT was then evaluated for presence of qualitative GCE and scored for SEBES using established criteria (see Supplemental Table I for definitions).^{8, 9} The amount of blood was assessed using the Hijdra sum score and the modified Fischer Scale (mFS).^{16, 17} Patients with qualitative GCE who experienced neurological deterioration or raised ICP within 72 hours were defined as suffering “clinical EBI”.

CSF was segmented from all CTs using a deep learning-based approach and then divided into ventricular and sulcal compartments (see Supplemental Figure I). Accuracy was assessed using the Dice similarity coefficient (DSC) and correlation of automated with manual CSF volumes on scans not used for training.¹⁸ Selective sulcal volume (SSV) was defined as the volume of CSF measured above the level of the lateral ventricles (see Figure 1). This was utilized as our primary quantitative biomarker of GCE, based on a prior study that determined its validity.¹⁴ Early SSV was then defined as the lowest SSV observed from all scans within 72 hours of ictus. One investigator reviewed all segmentation and SSV results and excluded any scans where this automated algorithm was unsuccessful.

Figure 1:

Results of automated CSF segmentation and ventricle/sulci separation to obtain selective sulcal volume (SSV, green) in SAH patients A) without GCE (SEBES of 0, SSV of 50.3 mL), and B) with GCE and sulcal effacement (SEBES of 4, SSV of 0.4 mL). Original CT top, segmented results bottom (sulcal volume at level of ventricles in fuchsia, and ventricles in red).

Statistical Analysis

Data analysis was performed using R version 3.6.1 (R Core Team, Vienna, Austria, 2019). Student’s t-test was used for continuous parametric variables; Kruskal-Wallis test for nonparametric variables; and chi-square or Fisher’s exact test for categorical variables. Receiver operating characteristic (ROC) curve analysis was used to evaluate early SSV as a biomarker for GCE using the package, pROC. The impact of age on SSV was evaluated using the correlation of admission SSV (excluding subjects with GCE) and subject age. We also analyzed changes in SSV from CTs taken prior to and after procedures that could influence edema, such as EVD placement or aneurysm repair. Univariate analysis was used to determine variables predictive of poor outcome. Variance Inflation Factor (VIF) was used to test for multicollinearity. Variables found significant on univariate analysis was included in the multivariable model. As both mFS and Hijdra score reflect the extent of bleeding, we chose to include only Hijdra in our final model. Separate models were made with and without an interaction term between age and measures of EBI. Model fit was compared using the Aikake Information Criteria (AIC, lower being better). Results of each model are reported as odds ratio (OR) and 95% confidence interval.

Results

Clinical and radiographic characteristics

320 patients were admitted with confirmed or presumed aSAH between January 2014 and March 2018. Of those, 244 met study inclusion criteria (see Supplemental Figure II for subject/scan flow). 90 (37%) had severe grade SAH on admission (WFNS 4–5). Hydrocephalus requiring EVD placement occurred in 162 (66%) and the ruptured aneurysm was treated by endovascular means in 126 (52%), surgically in 78 (32%), while 40 (16%) did not undergo a procedure, primarily due to poor clinical status. The median time from ictus to first CT was 6 hours (IQR 2–12), with 209 (86%) having admission CT within 24 hours of ictus. Of the 244 patients, 86 (35%) had evidence of EBI on admission CT per SEBES (i.e. score of 3–4) and 65 (27%) had qualitative GCE. An additional subset developed delayed-onset EBI as determined by high SEBES (n=34) or qualitative GCE (n=24) on subsequent scans within 72 hours of ictus. Therefore, a total incidence of EBI per SEBES was 120 (49%) and qualitative GCE was 89 (37%). Clinical EBI (radiographic GCE with associated symptomatic deterioration) occurred in 31 (13%). 25 (81%) of those with clinical EBI received osmotic therapies. Characteristics of subjects in relation to clinical EBI, qualitative GCE, and SEBES can be found in Table 1. Younger age was associated with all measures of EBI, while Hunt and Hess and WFNS scores were only higher in those with clinical EBI. Mortality and poor outcomes were also more likely in those with clinical EBI but not qualitative GCE or high SEBES.

Table 1.

Characteristics associated with EBI, as assessed using three definitions

	Clinical EBI (+) (n=31)	Clinical EBI (−) (n=211)	P Value	Qualitative GCE (+) (n=89)	Qualitative GCE (−) (n=155)	P Value	High SEBES (n=120)	Low SEBES (n=124)	P Value
Age	50 (12)	58 (15)	0.0011	48 (12)	64 (14)	<0.0001	50 (12)	65 (14)	<0.0001
Female	24 (77%)	151 (71%)	0.64	65 (73%)	112 (72%)	1	84 (70%)	93 (75%)	0.51
White Ethnicity	15 (48%)	141 (67%)	0.07	47 (53%)	111 (72%)	0.01	71 (59%)	87 (70%)	0.09
Hunt and Hess	4 (3–5)	3 (2–4)	0.001	3 (2–5)	3 (2–4)	0.33	3 (2–5)	3 (2–4)	0.33
WFNS	4 (2–5)	2 (1–4)	0.002	2 (2–4)	2 (1–4)	0.35	2 (2–4)	2 (1–4)	0.44
mFS	3 (3–4)	3 (1–4)	0.29	3 (1–4)	3 (2–4)	0.72	3 (1–4)	3 (2–4)	0.38
Hijdra Score	15 (10–18)	14 (7–20)	0.12	13 (6–17)	16 (7–21)	0.19	13 (6–17)	16 (8–21)	0.41
DCI	5 (16%)	45 (21%)	0.67	19 (21%)	31 (20%)	0.93	26 (22%)	24 (19%)	0.77
LOC	19 (61%)	95 (45%)	0.09	48 (54%)	67 (43%)	0.10	62 (51%)	53 (43%)	0.16
Anterior Aneurysm	22 (71%)	161 (76%)	0.52	68 (76%)	116 (75%)	0.78	94 (78%)	90 (73%)	0.30
Posterior Aneurysm	3 (10%)	34 (16%)	0.35	11 (12%)	26 (17%)	0.35	16 (13%)	21 (17%)	0.43
Aneurysm pattern without angiogram	6 (19%)	16 (8%)	0.03	10 (11%)	13 (8%)	0.46	10 (8%)	13 (10%)	0.57
Hydrocephalus requiring EVD	26 (84%)	135 (64%)	0.05	61 (69%)	101 (65%)	0.69	83 (69%)	79 (64%)	0.44
Aneurysm treatment modality
Clipping	13 (42%)	65 (31%)	0.003 (0.04)*	39 (44%)	39 (25%)	0.004 (0.001)*	49 (41%)	29 (23%)	0.01 (0.009)*
Coiling	8 (26%)	117 (55%)		34 (38%)	92 (59%)		54 (45%)	72 (58%)
None	10 (32%)	29 (14%)		16 (18%)	24 (15%)		17 (14%)	23 (19%)
Mortality	14 (45%)	48 (23%)	0.02	21 (24%)	42 (27%)	0.65	25 (21%)	38 (31%)	0.11
Poor Outcome	20 (65%)	92 (43%)	0.05	40 (45%)	73 (47%)	0.85	53 (44%)	60 (48%)	0.59
Early SSV	0.10 (0.02 – 0.38)	4.17 (0.77 – 13.86)	<0.0001	0.18 (0.04 – 1.12)	8.21 (2.79 – 20.79)	<0.0001	0.42 (0.06 – 2.05)	11.14 (4.07 – 25.62)	<0.0001
Admission SSV	1.39 (0.33 – 5.55)	13.53 (2.81 – 29.61)	<0.0001	1.67 (0.38 – 4.30)	20.59 (9.59 – 37.35)	<0.0001	2.56 (0.65 – 8.71)	24.73 (11.64 – 41.80)	<0.0001
Admission SEBES	4 (4–4)	2 (0–4)	<0.0001	4 (4–4)	1 (0–2)	<0.0001	4 (4–4)	0(0–2)	<0.0001

Values are displayed as mean (SD), median (IQR), count (%)

^*P-value aneurysm treatment modality displayed as: among all three groups (post-hoc analysis between clipping and coiling)

We obtained automated SSV from 654 CT scans within 72 hours of ictus. Accuracy of automated CSF segmentation was high, with a mean DSC of 0.82±0.11. The correlation between automated and manually measured volumes was 0.99 (p=0.009). In 156 (64%) subjects, SSV decreased on repeat CT within 72 hours of ictus, resulting in a median early SSV (lowest from scans within 72 hours) of 3.0 mL (IQR 0.3 – 11.9). The median time to reach SSV nadir was 18 hours (IQR 8 – 38). EVD placement was not associated with improvement in edema (median change in SSV – measured between the last scan within 72 hours of ictus after EVD placement and minimum SSV prior to EVD placement – was negligible at −0.05-ml, IQR −5.4 to 0.2-ml in 100 patients with scans pre- and post-EVD). Only two SAH patients with low SSV prior to EVD saw a significant (5-ml or greater) improvement in SSV after EVD. Similarly, there was no difference in the change in SSV after aneurysm repair procedures (median change in SSV from pre-procedure to lowest value on post-procedure CT within 72 hours of ictus was −1.2 ml in those with endovascular repair compared with −1.1-ml with surgical repair, p=0.94).

Lower early SSV was correlated with higher WFNS (ρ = −0.297) and Hunt and Hess scores (ρ = −0.310, both p<0.0001) and with loss of consciousness at ictus (p<0.0001). Early SSV was strongly negatively correlated to maximum SEBES (ρ = −0.72; p < 0.0001; see Figure 2A) and those with qualitative GCE and clinical EBI had significantly lower early SSV (Supplemental Figure III). Each mL decrease in SSV from admission was associated with qualitative GCE (OR 1.90; 95% CI 1.08–3.36), high SEBES (OR 1.95; 95% CI 1.15–3.32), and clinical EBI (OR 2.60; 95% CI 1.02–6.60). ROC analysis indicated that early SSV was an accurate marker for qualitative GCE (AUC = 0.90; 95% CI 0.86–0.94; see Figure 2B) as well as GCE and clinical EBI (see Supplemental Figure IV). A cutoff of SSV below 5.2 mL within 72 hours provided a sensitivity of 97% to diagnose qualitative GCE (with 60% specificity). Using this cutoff, 148 (61%) of SAH subjects would be categorized as having GCE based on low SSV. Variables associated with low SSV are shown in Supplemental Table II.

Figure 2.

A) Early SSV in relation to maximum SEBES within 72 hours of ictus across all subjects (correlation, ρ = −0.72); B) ROC curve for the assessment of GCE using early SSV; C) Relationship between early SSV and age. Numbers above each boxplot show the percentage of patients in each age group who were positive for GCE.

Age, WFNS, and loss of consciousness at ictus were independent predictors of developing edema, as defined by low SSV within 72 hours, while hydrocephalus (requiring EVD placement) or type of aneurysm repair, were not (full results of multivariable model in Supplemental Table III).

Age dependency of GCE, SEBES, and SSV

Younger patients were more likely to develop qualitative GCE (OR 2.37 per 10 years, 95% CI 1.79–3.11; p<0.0001). Maximum SEBES was also negatively correlated with age (ρ = −0.555; p < 0.0001; Supplemental Figure V). Similarly, younger patients had lower admission SSV, even after removing those with qualitative GCE whose volumes might be depressed by injury (r = 0.210; p = 0.0001; Supplemental Figure VI). With age divided into deciles, age was positively correlated with early SSV (ρ = 0.404; p < 0.0001) and younger patients were more likely to be diagnosed with GCE (Figure 2C).

Prognostic impact of early SSV on outcome

Out of 244 SAH patients, 113 (46 %) had a poor functional status at hospital discharge (Table 2). Clinical EBI and lower early SSV (but not admission SSV) were both associated with poor outcome on univariate analysis (OR 2.35; 95% CI 1.09–5.32 and 1.03 per ml; 95% CI 1.01–1.06 respectively) while qualitative GCE and SEBES were not. Developing DCI was also not associated with worse outcomes.

Table 2.

Characteristics associated with functional outcome at discharge

	Total (n=244)	Poor Outcome (n=113)	Good Outcome (n=131)	P-value
Age	57 (15)	62 (15)	54 (15)	<0.0001
Female	177 (73%)	84 (74%)	93 (71%)	0.66
White Ethnicity	158 (65%)	77 (68%)	81 (62%)	0.37
Hunt and Hess	3 (2–4)	4(3–5)	2(2–3)	<0.0001
WFNS	2 (1–3)	4(2–5)	2(1–2)	<0.0001
mFS	3 (2–4)	3(2–4)	2(1–3)	0.0002
Hijdra Score	14 (7–20)	17 (10–21)	12 (5–18)	<0.0001
DCI	50 (20%)	22 (19%)	28 (21%)	0.83
LOC	115 (47%)	72 (65%)	43 (33%)	<0.0001
Anterior Aneurysm	184 (75%)	73 (65%)	111 (85%)	0.0003
Posterior Aneurysm	37 (15%)	17 (15%)	20 (15%)	0.96
Aneurysm pattern without angiogram	23 (9%)	23 (20%)	0 (0%)	<0.0001
Hydrocephalus requiring EVD	162 (66%)	80 (71%)	82 (63%)	0.22
Aneurysm Treatment Modality
Clipping	78 (32%)	26 (23%)	52 (40%)	<0.0001 (0.5)*
Coiling	126 (52%)	49 (43%)	77 (59%)
None	40 (16%)	38 (34%)	2 (2%)
Time to initial CT (hr)	5.8 (1.9–11.7)	5.0 (1.4–9.3)	6.7 (2.3–15.8)	0.46
Clinical EBI	31 (13%)	20 (18%)	11 (8%)	0.05
Qualitative GCE within 72 hours	89 (37%)	40 (35%)	49 (37%)	0.85
Max SEBES within 72 hours	2 (1–4)	2(1–4)	3(0–4)	0.06
Early SSV	3.01 (0.34–11.85)	1.64 (0.12–8.94)	4.20 (0.74–18.24)	0.001
Qualitative GCE on initial CT	65 (27%)	30 (27%)	35 (27%)	1
SEBES on initial CT	2(0–4)	2 (0–4)	2 (0–4)	0.55
SSV on initial CT	10.70 (2.01–28.23)	9.23 (1.33–27.00)	11.37 (2.74–28.48)	0.21

Values are displayed as mean (SD), median (IQR), count (%)

^*P-value aneurysm treatment modality displayed as: among all three groups (post-hoc analysis between clipping and coiling)

The multivariable model (Table 3) found that early SSV, along with higher age and WFNS, were associated with poor outcome; for each mL reduction in SSV, the odds for poor outcome increased 3% (OR 1.03, 95% CI 1.01–1.06). Adjustment for hydrocephalus and aneurysm therapy (clipping vs. coiling) did not affect these results. After introducing the interaction of age with SSV (model 2), model fit increased (AIC 265.93 vs. 269.35) and the impact of SSV on outcome was more pronounced (OR 1.21, 95% CI 1.05 – 1.43). Figure 3 shows how the effect of early SSV on outcome varies with age, illustrating that early SSV has a stronger effect in younger patients and remains predictive only in individuals less than 70 years of age. Although clinical EBI was associated with outcome on univariate analysis, neither clinical EBI, SEBES, nor qualitative GCE showed an independent effect on outcome, even after adjustment for their interactions with age (Supplemental Table IV). In contrast, when early SSV was dichotomized (with GCE defined as SSV < 5.2 mL), low SSV was independently predictive of poor outcome, but only for individuals younger than 58 years of age (see Supplemental Figure VII).

Table 3.

Model for the prediction of poor outcome at discharge using early SSV. Model 2 includes the interaction between age and early SSV whereas Model 1 does not.

	Model 1			Model 2
	B	OR (95 CI)	P-Value	B	OR (95 CI)	P-value
Age (per 1-year increase)	0.042	1.04 (1.02 – 1.07)	<0.001	0.028	1.03 (1.0 – 1.06)	0.04
WFNS* (per 1-point increase)	0.570	1.77 (1.38 – 2.30)	<0.001	0.554	1.74 (1.36 – 2.27)	<0.001
LOC	−0.039	0.96 (0.45 – 1.97)	0.92	−0.025	0.98 (0.46 – 2.00)	0.95
Hijdra score (per 1-point increase)	0.016	1.02 (0.97 – 1.06)	0.46	0.018	1.02 (0.98 – 1.06)	0.42
Early SSV (per 1 mL decrease)	0.039	1.03 (1.01 – 1.06)	0.02	0.187	1.21 (1.05 – 1.43)	0.02
Interaction of age and early SSV	NA					0.04

^*Due to high multicollinearity between Hunt and Hess and WFNS (VIF = 11.1 and 11.2 respectively), we chose to only include WFNS

Figure 3.

Marginal effects plot depicting how the estimated coefficient for the influence of early SSV on outcome changes with age (interaction p=0.04).

Discussion

The key findings of our study are as follows: 1) SSV (i.e. sulcal volume above the ventricles) can be accurately measured by our novel automated imaging algorithm in a large cohort of aSAH patients; 2) an SSV cutoff below 5-ml accurately identifies GCE in SAH patients; 3) EBI as identified by qualitative GCE, high SEBES, and low SSV was observed in 27, 35, and 40% (respectively) of patients on admission, but also developed in another 9, 14, and 21% (respectively) of patients in delayed fashion (within 72 hours); 4) Early SSV (i.e. the lowest SSV within 72 hours of ictus) was an independent predictor of poor outcome in our aSAH cohort, while qualitative GCE, high SEBES, and admission SSV were not; and 5) Early SSV had its strongest impact on functional outcome among younger SAH patients. These results suggest that automated quantification of SSV may provide a widely accessible and reliable biomarker for EBI after SAH.

It is now accepted that EBI is a critical but under-studied contributor to outcome after SAH. Recently, two studies found that GCE, rather than DCI, is the major cause of mortality in patients with aSAH.^{12, 19} We also found that DCI was not associated with outcome. Of the many factors associated with poor cognitive and functional outcome, GCE may be the only one potentially amenable to treatment.²⁰ However, one of the major obstacles to studying the spectrum of EBI is the lack of reliable means of quantifying it. The presence of GCE is a core radiographic manifestation of EBI and is typically defined based on effacement of hemispheric sulci.⁹ However, this qualitative assessment of GCE relies on subjective rating and categorizes only whether severe edema is present or absent. To address this limitation, a semi-quantitative method for GCE has been developed – using a scoring schema known as SEBES.⁸ Though SEBES has been an important advance in the identification of EBI in SAH patients, this system (like qualitative GCE) relies on manual inspection and subjective adjudication of sulcal effacement, limiting its applicability to the kind of large cohort studies required to understand the underlying biology and impact of EBI in SAH patients.

Our automated technique for the quantification of GCE is built upon a previously published neural network-based method for quantification of CSF volume following ischemic stroke.^{21, 22} We chose to focus on quantifying SSV for two reasons: 1) Choi and colleagues previously demonstrated that SSV could accurately quantify the severity of GCE after SAH;¹⁴ and 2) Eibach and colleagues recently found that sulcal effacement above the lateral ventricles was more indicative of GCE than loss of sulci at or below the level of the lateral ventricles.²³ We employed an automated segmentation algorithm to extract these CSF regions from serial CTs, previously performed manually on a single admission scan.¹⁴ These innovations allowed us to measure SSV on hundreds of serial CT scans with minimal manual review. This automated approach will facilitate the objective assessment of GCE in large cohorts of SAH patients, enabling studies to elucidate its underlying biology, document its impact on patient outcome, and eventually examine the efficacy of EBI-directed therapeutic interventions.

Moreover, it provides a dynamic means of assessing the evolution of edema-related injury over time. Measurement of SSV allows us to move beyond simply labeling GCE as present or absent, based on a single threshold. We demonstrated that this method is sensitive to reductions in sulcal volume over the first few days after SAH, as GCE and EBI often progress. One recent study highlighted that GCE (using SEBES) evolves over time, is often greater one day after admission, and resolves at different rates.²⁴ The time course of edema was found to be prognostically significant. We similarly found that admission SSV and SEBES were not associated with outcome, while lowest SSV within 72 hours was. We also demonstrated that early interventions, such as treatment of hydrocephalus and aneurysm repair, do not substantially alter the trajectory and severity of edema after SAH. While osmotic therapies were given to most patients with symptomatic GCE, this was after development of edema and would be expected to increase already low SSV and not influence our assessment of edema by the lowest measured SSV.

We propose a model of EBI in which SSV quantification captures a broader spectrum of severity than qualitative grading of GCE. Reduced SSV was found in 61% of subjects while qualitative GCE was seen in only a subset of those with measurable edema (36%). Far fewer of both these groups experienced neurologic deterioration associated with edema (i.e. clinical EBI); nonetheless, this seems to represent a more severe sub-group with higher mortality. Further studies will have to corroborate the significance of clinical EBI and its relationship to the broader group with edema as defined by reduced SSV.

Several prior studies have found that different measures of EBI, from qualitative GCE, higher SEBES, and lower SSV, were associated with worse outcomes after SAH.^{8, 9, 14, 23} Quantifying edema using SSV may increase the power to extract meaningful signals of how the spectrum of EBI influences outcomes. Furthermore, no prior studies have specifically evaluated the impact of age on EBI, not only as a confounder (i.e. younger individuals are likely to have lower sulcal volumes, irrespective and prior to injury), but also whether an interaction between age and EBI exists. Not only did we validate that quantitative GCE (using SSV) was associated with outcomes (while, in our cohort, qualitative assessment of GCE and use of SEBES were not), moreover, we found but that this association was age-dependent, meaning that the effect was not the same for young and for older patients. Beyond the age of 70, edema (assessed by lower SSV) did not seem to impact outcomes. The evidence that GCE is more relevant for younger aSAH patients is not novel: Claassen et al. reported that GCE was only predictive of poor outcome in individuals younger than 58 years of age,⁹ while Eibach et al. reported that SEBES was only a reliable predictor of outcome in patients younger than 60 years of age.²³ Our quantitative approach provides greater insights into the mechanisms of this interaction: younger SAH patients have less pre-existing CSF ‘reserve’ and are more likely to become symptomatic and have their outcomes influenced by the reduction in CSF volumes effected by EBI-related edema. Older patients generally have greater atrophy and are better able to tolerate swelling. It may also be that any impact of reduced SSV (from edema) is negated by the adverse consequences on recovery resulting from aging, as those with more sulcal volume (i.e. greater atrophy) are at higher risk for poor outcomes from edema-independent mechanisms. These findings suggest that there is a delicate balance between sulcal volume and SAH-related injury that shifts with age, providing one hypothesis for why GCE only predicts poor outcome in younger aSAH patients.

Limitations

The retrospective nature of this study resulted in real-world variation between the timing and number of scans performed per patient. This may introduce bias, as patients who remained clinically stable likely would not receive as many follow-up CTs. In contrast, it is likely that those with severe edema would be more likely to have imaging and so would be captured. We blinded investigators to patient data while evaluating qualitative GCE and SEBES to limit rater bias. We only measured sulcal volume by segmentation of CSF-filled sulci, which could underestimate total volume if other sulci were filled with blood. However, in our review of these SAH scans, the quantity of sulcal blood above the ventricles was minimal and unlikely to substantially alter assessment of SSV. Nonetheless, we are now developing segmentation algorithms that will extract not only CSF but also sulcal blood as well as blood in other brain compartments. Lastly, our study is limited to a single center investigation and only evaluated short-term outcome after aSAH. We used functional status at hospital discharge, as we had limited follow-up data on this cohort. However, another study of SAH patients with serial follow-up found that there is generally at-most a one-point improvement in mRS between discharge and 6-month follow-up.²⁵ We assigned poor outcome as mRS of 4 or greater at discharge, which therefore likely corresponds to at best a mRS of 3 or more on follow-up, generally considered poor functional recovery after SAH.²⁶ Furthermore, prior SAH studies have found similar effects of EBI and edema on outcomes when using discharge mRS as with application of functional outcomes at follow-up.⁸ Nevertheless, prospective external validation of this algorithm in prediction of long-term outcomes across multiple institutions is needed to address these limitations.

Conclusion

Reduced early SSV is the first successful quantitative measure of edema in SAH patients, which provides synergy with both of the widely used qualitative biomarkers (qualitative GCE and SEBES). This is desirable as it allows for a more complete representation of the full spectrum of EBI in this patient population. GCE quantification using early SSV is a novel, objective measure that provides an opportunity for earlier recognition and consistent tracking of brain edema after SAH.

Non-standard Abbreviations and Acronyms:

(aSAH)

Aneurysmal Subarachnoid hemorrhage

(CSF)

Cerebrospinal fluid

(DSC)

Dice Similarity Coefficient

(EBI)

Early brain injury

(EVD)

External ventricular drain

(GCE)

Global cerebral edema

(ICP)

Intracranial pressure

(LOC)

Loss of consciousness

(mRS)

Modified Rankin Scale

(SSV)

Selective sulcal volume

(SEBES)

Subarachnoid hemorrhage early brain edema score

(WFNS)

World Federation of Neurosurgical Societies’