Background and Purpose— Cerebral edema (CED) develops in the hours to days after stroke; the resulting increase in brain volume may lead to midline shift (MLS) and neurological deterioration. The time course and implications of edema formation are not well characterized across the spectrum of stroke. We analyzed displacement of cerebrospinal fluid (ΔCSF) as a dynamic quantitative imaging biomarker of edema formation. Methods— We selected subjects enrolled in a stroke cohort study who presented within 6 hours of onset and had baseline and ≥1 follow-up brain computed tomography scans available. We applied a neural network-based algorithm to quantify hemispheric CSF volume at each imaging time point and modeled CSF trajectory over time (using a piecewise linear mixed-effects model). We evaluated ΔCSF within the first 24 hours as an early biomarker of CED (defined as developing MLS on computed tomography beyond 24 hours) and poor outcome (modified Rankin Scale score, 3–6). Results— We had serial imaging in 738 subjects with stroke, of whom 91 (13%) developed CED with MLS. Age did not differ (69 versus 70 years), but baseline National Institutes of Health Stroke Scale was higher (16 versus 7) and baseline CSF volume lower (132 versus 161 mL, both P <0.001) in those with CED. ΔCSF was faster in those developing MLS, with the majority seen by 24 hours (36% versus 11% or 2.4 versus 0.8 mL/h; P <0.0001). Risk of CED almost doubled for every 10% ΔCSF within 24 hours (odds ratio, 1.76 [95% CI, 1.46–2.14]), adjusting for age, glucose, and National Institutes of Health Stroke Scale. Risk of neurological deterioration (1.6-point increase in National Institutes of Health Stroke Scale at 24 hours) and poor outcome (adjusted odds ratio, 1.34 [95% CI, 1.15–1.56]) was also greater for every 10% increase in ΔCSF. Conclusions— CSF volumetrics provides quantitative evaluation of early edema formation. ΔCSF from baseline to 24-hour computed tomography is a promising early biomarker for the development of MLS and worse neurological outcome.

Background and Purpose— Volumes of hemorrhage and perihematomal edema (PHE) are well-established biomarkers of primary and secondary injury, respectively, in spontaneous intracerebral hemorrhage. An automated imaging pipeline capable of accurately and rapidly quantifying these biomarkers would facilitate large cohort studies evaluating underlying mechanisms of injury. Methods— Regions of hemorrhage and PHE were manually delineated on computed tomography scans of patients enrolled in 2 intracerebral hemorrhage studies. Manual ground-truth masks from the first cohort were used to train a fully convolutional neural network to segment images into hemorrhage and PHE. The primary outcome was automated-versus-human concordance in hemorrhage and PHE volumes. The secondary outcome was voxel-by-voxel overlap of segmentations, quantified by the Dice similarity coefficient (DSC). Algorithm performance was validated on 84 scans from the second study. Results— Two hundred twenty-four scans from 124 patients with supratentorial intracerebral hemorrhage were used for algorithm derivation. Median volumes were 18 mL (interquartile range, 8–43) for hemorrhage and 12 mL (interquartile range, 5–30) for PHE. Concordance was excellent (0.96) for automated quantification of hemorrhage and good (0.81) for PHE, with DSC of 0.90 (interquartile range, 0.85–0.93) and 0.54 (0.39–0.65), respectively. External validation confirmed algorithm accuracy for hemorrhage (concordance 0.98, DSC 0.90) and PHE (concordance 0.90, DSC 0.55). This was comparable with the consistency observed between 2 human raters (DSC 0.90 for hemorrhage, 0.57 for PHE). Conclusions— We have developed a deep learning-based imaging algorithm capable of accurately measuring hemorrhage and PHE volumes. Rapid and consistent automated biomarker quantification may accelerate powerful and precise studies of disease biology in large cohorts of intracerebral hemorrhage patients.

As swelling occurs, CSF is preferentially displaced from the ischemic hemisphere. The ratio of CSF volume in the stroke-affected hemisphere to that in the contralateral hemisphere may quantify the progression of cerebral edema. We automatically segmented CSF from 1,875 routine CTs performed within 96 hours of stroke onset in 924 participants of a stroke cohort study. In 737 subjects with follow-up imaging beyond 24-hours, edema severity was classified as affecting less than one-third of the hemisphere (CED-1), large hemispheric infarction (LHI, over one-third the hemisphere), without midline shift (CED-2) or with midline shift (CED-3). Malignant edema was LHI resulting in deterioration, requiring osmotic therapy, surgery, or resulting in death. Hemispheric CSF ratio was lower on baseline CT in those with LHI (0.91 vs. 0.97, p < 0.0001) and decreased more rapidly in those with LHI who developed midline shift (0.01 per hour for CED-3 vs. 0.004/hour CED-2). The ratio at 24-hours was lower in those with midline shift (0.41, IQR 0.30–0.57 vs. 0.66, 0.56–0.81 for CED-2). A ratio below 0.50 provided 90% sensitivity, 82% specificity for predicting malignant edema among those with LHI (AUC 0.91, 0.85–0.96). This suggests that the hemispheric CSF ratio may provide an accessible early biomarker of edema severity.

Introduction: Malignant cerebral edema develops in a small subset of patients with hemispheric strokes, precipitating deterioration and death if decompressive hemicraniectomy (DHC) is not performed in a timely manner. Predicting which stroke patients will develop malignant edema is imprecise based on clinical data alone. Head computed tomography (CT) imaging is often performed at baseline and 24-h. We determined the incremental value of incorporating imaging-derived features from serial CTs to enhance prediction of malignant edema. Methods: We identified hemispheric stroke patients at three sites with NIHSS ≥ 7 who had baseline as well as 24-h clinical and CT imaging data. We extracted quantitative imaging features from baseline and follow-up CTs, including CSF volume, intracranial reserve (CSF/cranial volume), as well as midline shift (MLS) and infarct-related hypodensity volume. Potentially lethal malignant edema was defined as requiring DHC or dying with MLS over 5-mm. We built machine-learning models using logistic regression first with baseline data and then adding 24-h data including reduction in CSF volume (ΔCSF). Model performance was evaluated with cross-validation using metrics of recall (sensitivity), precision (predictive value), as well as area under receiver-operating-characteristic and precision-recall curves (AUROC, AUPRC). Results: Twenty of 361 patients (6%) died or underwent DHC. Baseline clinical variables alone had recall of 60% with low precision (7%), AUROC 0.59, AUPRC 0.15. Adding baseline intracranial reserve improved recall to 80% and AUROC to 0.82 but precision remained only 16% (AUPRC 0.28). Incorporating ΔCSF improved AUPRC to 0.53 (AUROC 0.91) while all imaging features further improved prediction (recall 90%, precision 38%, AUROC 0.96, AUPRC 0.66). Conclusion: Incorporating quantitative CT-based imaging features from baseline and 24-h CT enhances identification of patients with malignant edema needing DHC. Further refinements and external validation of such imagingbased machine-learning models are required.

Quantifying the extent and evolution of cerebral edema developing after stroke is an important but challenging goal. Lesional net water uptake (NWU) is a promising CT-based biomarker of edema, but its measurement requires manually delineating infarcted tissue and mirrored regions in the contralateral hemisphere. We implement an imaging pipeline capable of automatically segmenting the infarct region and calculating NWU from both baseline and follow-up CTs of large-vessel occlusion (LVO) patients. Infarct core is extracted from CT perfusion images using a deconvolution algorithm while infarcts on follow-up CTs were segmented from non-contrast CT (NCCT) using a deep-learning algorithm. These infarct masks were flipped along the brain midline to generate mirrored regions in the contralateral hemisphere of NCCT; NWU was calculated as one minus the ratio of densities between regions, removing voxels segmented as CSF and with HU outside thresholds of 20–80 (normal hemisphere and baseline CT) and 0–40 (infarct region on follow-up). Automated results were compared with those obtained using manually-drawn infarcts and an ASPECTS region-of-interest based method that samples densities within the infarct and normal hemisphere, using intraclass correlation coefficient (ρ). This was tested on serial CTs from 55 patients with anterior circulation LVO (including 66 follow-up CTs). Baseline NWU using automated core was 4.3% (IQR 2.6–7.3) and correlated with manual measurement (ρ = 0.80, p &lt; 0.0001) and ASPECTS (r = −0.60, p = 0.0001). Automatically segmented infarct volumes (median 110-ml) correlated to manually-drawn volumes (ρ = 0.96, p &lt; 0.0001) with median Dice similarity coefficient of 0.83 (IQR 0.72–0.90). Automated NWU was 24.6% (IQR 20–27) and highly correlated to NWU from manually-drawn infarcts (ρ = 0.98) and the sampling-based method (ρ = 0.68, both p &lt; 0.0001). We conclude that this automated imaging pipeline is able to accurately quantify region of infarction and NWU from serial CTs and could be leveraged to study the evolution and impact of edema in large cohorts of stroke patients.