Perihematomal diffusion restriction as a common finding in large intracerebral hemorrhages in the hyperacute phase

Purpose There is growing evidence that a perihematomal area of restricted diffusion (PDR) exists in intraparenchymal hemorrhages (IPH) within 1 week of symptom onset (SO). Here, we study characteristics and the clinical impact of the PDR in patients with hyperacute (≤ 6 hours from SO) IPH by means of apparent diffusion coefficient (ADC). Methods This monocentric, retrospective study includes 83 patients with first-ever primary IPH from 09/2002-10/2015. 3D volumetric segmentation was performed for the IPH, PDR, and perihematomal edema (PHE) on fluid-attenuated inversion recovery, T2*/susceptibility weighted images, and ADC images. Results A PDR was seen in 56/83 patients (67.5%) presenting with hyperacute IPH. Multivariate logistic regression analysis revealed every 10-year increase of age (HR 1.929, 95% CI 1.047–3.552, P = .035) and male gender (HR 5.672, 95% CI 1.038–30.992, P = .045) as significant predictors of the presence of a PDR, but not IPH size, IPH location, nor National Institutes of Health Stroke Scale Score (NIHSS) at admission. We found no difference in NIHSS at discharge, hematoma removal, or mortality rate in PDR-positive patients. ADC values of the PDR show a step-wise normalization with increasing time from SO. Conclusions Occurrence of a PDR is a common finding in supratentorial hyperacute IPH, but shows no adverse short-term clinical impact. It may represent transient oligemic and metabolic changes.

Introduction Intracerebral hemorrhage or intraparenchymal hemorrhage (IPH) represents a second cause of stroke with high blood pressure being the main risk factor [1,2]. On magnetic resonance imaging (MRI), 5 different stages of hematoma formation are recognized: hyperacute ( 6 hours from symptom onset, SO), acute (> 6 to 72 hours), early subacute (> 3 to 7 days), late subacute (> 1 to 2 weeks), and chronic (2 weeks to years) [3]. Due to lysis of red blood cells and changes in the oxygenation state of hemoglobin, all stages show a characteristic pattern on MRI [4,5].
Here, we aim to determine the morphology and the clinical impact of the PHE region in hyperacute IPH patients by means of ADC. We hypothesize that ADC hypointensities are frequently found within the PHE and that this finding is associated with a worse clinical status and outcome.

Study cohort
This single center, retrospective study was conducted in compliance with the local ethics committee (Ethik-Kommission der Ä rztekammer Hamburg, WF-023/15) with a waiver of informed consent. We reviewed all patients ! 18 years in whom diagnosis of spontaneous first-ever IPH was made after an MRI was performed due to suspected acute ischemic stroke from 09/2002 to 10/2015 6 hours from SO (hyperacute IPH). Patients with hemorrhagic conversion of acute ischemic stroke, traumatic IPH, underlying brain malignancy, and secondary causes of IPH (e.g. arteriovenous malformations and venous sinus thrombosis-since pathophysiology includes vessel occlusion and therefore a higher likelihood of ischemia) were excluded.
In total, 83 patients (54 males with a median age of 64 (interquartile range (IQR) 54.75-73.25) years and 29 females with a median age of 71 (IQR 58-80.5) years) met the inclusion criteria. IPH was classified as deep if the thalamus, basal ganglia, and/or internal capsule were suspected as bleeding origin. In a second step, electronic chart review of eligible cases was performed and medication, NIHSS on admission and discharge as assessed by the treating physician were recorded.

Image analysis
Image Analysis was performed with Analyze Software System 11.0 (Biomedical Imaging Resource, Mayo Clinic, Rochester, MN, USA) by a single reader blinded to the patients status (D. F.) [15]. For this purpose, all sequences (FLAIR, ADC, and T2 Ã or SWI) were spatially coregistered to one another using three-dimensional (3D) voxel registration module. If necessary, the registration was completed using manual 3D registration. Hence segmented regions of interest (ROIs) were transferable between all sequences. First, the parenchymal hematoma was contoured manually on all T2 Ã or SWI images. Since T2 Ã and SWI images have been shown to overestimate the size of the hematoma by approximately 20%, the co-registered FLAIR was used to precisely define the borders of the clot (IPH ROI) [16]. Afterwards, the whole lesion, defined by the borders of the PHE, was segmented on FLAIR images (lesion ROI), Fig 1. IPH volume directly corresponded to the IPH ROI (in cc), whereas for calculation of PHE volume IPH ROI was subtracted from lesion ROI (lesion ROI-IPH ROI = PHE ROI, in cc).
In a majority of cases, we found a rim of restricted diffusion adjacent to the outer border of the PHE. This perihematomal rim (PDR) was defined as hyperintense area on b1000 images and hypointense on the corresponding ADC map applying a threshold of 1.5 standard deviations of the median ADC value in a manually corrected mirror region (Fig 1). In cases a PDR was present, it was segmented on ADC images. For calculation of relative ADC values (rADC), median absolute ADC values of IPH, PHE, and PDR were each divided by median ADC values of the manually drawn corresponding contralateral mirror region. Two segmentation examples are shown in Fig 2. Accuracy of lesion masks was evaluated by 2 independent readers (S. S. and T. S. with 10 and 4 years of experience in Neuroradiology, respectively). For IPH, edema, and PDR ROI, P25 (first quartile), mean, median, and P75 (third quartile) ADC values were read and volumes were computed. Follow-up images were visually assessed by S. S. and T. S. in consensus and patients were classified into 2 groups, based on whether the IPH decreased or increased in size during the first 10 days after SO.

Statistical analysis
Pearson's correlation was used to compare IPH and PDR volumes. The differences in age, IPH volume, NIHSS at admission and at discharge among PDR-positive and PDR-negative patients, differences in PDR volumes by location, PDR/IPH volumes, and absolute ADC values in patients with and without IPH growth within 10 days after SO, in patients with and without hematoma removal, and patients who died were compared by Mann-Whitney U test. Pearson's chi-squared test was performed to test gender frequency, time from symptom onset to MRI, antiplatelet therapy status, and the frequency of lobar and deep IPH in PDR-positive and PDR-negative patients. Fisher's Exact test was used to determine differences of the frequency of a PDR in supra-and infratentorial locations, hematoma removal rate, mortality rate, anticoagulation therapy status, and the frequency of IPH volume increase within 10 days after SO. Differences in PDR size and time from symptom onset to MRI were determined by using the Kruskal-Wallis test.
For multivariate analysis, a logistic regression was performed including gender, age, IPH location, time from SO to MRI, NIHSS at admission, and log IPH volume. A P value < .050 was considered significant. Data collection was performed using IBM SPSS Statistics1, version 20 (IBM1 2011, Armonk, NY, USA). If not otherwise indicated, data are given as median and IQR (in parenthesis).

Descriptive statistics
Baseline patient characteristics are given in Table 2. IPH location was supratentorial in 81 cases (66 deep, 15 lobar) and infratentorial in 2 cases (1 cerebellar, 1 pontine). MRI was performed after 2.0 (1.0-2.75) hours from SO. In 15 patients, it was acquired within 6 hours from SO (assured last seen well-time below 6 hours), but exact SO was not observed ("wake-up IPH", only supratentorial IPH).

Outcome measurements
Median hospital stay was 16 (9-21) days. Median NIHSS for supra-and infratentorial IPH at initial presentation was 11 (8-15) and 4.5 points and at discharge from hospital 7 (4-11) and 3 points, respectively, without a difference in NIHSS between patients with and without PDR (initial NIHSS: U = 560.5, P = .072; NIHSS at discharge: U = 561.5, P = .872). Cases with higher IPH volumes showed significantly higher NIHSS at admission (r = 0.302, P = .006) and at discharge (r = 0.293, P = .012). Of the 41 patients who underwent follow-up imaging within 10 days after SO, 4 showed hematoma enlargement that was neither associated with presence of a PDR (P = .129). We found no difference in absolute IPH, PHE, and PDR ADC values in patients with and without IPH growth (U > 7.5, P > .571). 4 patients underwent hematoma evacuation due to life threatening mass effect with consecutive impairment of cerebrospinal fluid flow immediately following admission. Those patients neither showed a significantly higher initial PDR (U = 85, P = .570) nor IPH volume (U = 77, P = .087) compared to conservatively treated cases, and we found no difference in ADC distribution of the PDR between both groups (U > 95.5, P > .794). 7/83 patients died after 18 (3tabl-27.5) days from SO regardless of the presence of a PDR (P = .711). Deceased PDR-positive patients had higher PDR and IPH volumes than PDR-negative cases (U = 104.5, P = .098 and U = 68, P < .001, respectively), but there were also no differences in absolute ADC values (U > 103.5, P > .093).

Discussion
ADC is a measure of the diffusibility of water molecules within a tissue and is determined from DWI. Low ADC values correspond to restricted water diffusion and are considered to be caused by cell swelling or increased cellular density [17]. In hyperacute IPH patients, evidence for diffusion restriction in the periclot area was shown in 3 studies: Kidwell et al. first described a rim of decreased ADC values in a subset of 3/12 patients, but did not acquire FLAIR sequences. Therefore, it becomes not entirely clear whether the rim is part of the edema or is entirely located outside the edema [13]. Schellinger et al. reported 7/32 patients with lower ADC values/cytotoxic edema within a ROI of 1 cm surrounding the IPH compared to the healthy hemisphere [14]. The most recent study by Stösser et al. showed a PDR in 88/172 (51.2%) IPH patients within 24 hours of symptom onset, but they did not separately examine the subgroup of patients in a hyperacute or 6-hour time window [12]. The MRI pattern of low ADC without corresponding hyperintensity on FLAIR is wellknown from ischemic stroke patients in the first few hours with the tissue being hypoperfused,  but potentially salvageable from ischemia. But do the same mechanisms causing brain damage in ischemic stroke occur in the PDR? Since perfusion deficits were shown to occur in the ipsilateral hemisphere rather than within the perihematomal area (Kidwell et al.) while no correlation between cytotoxic edema and prolongation of mean transit time was found (Schellinger et al.), both authors conclude that the rim/ROI of decreased ADC primarily represents temporary depression of regional neuronal metabolism rather than true ischemia. Comparable conclusions were drawn from perfusion/diffusion studies of acute and hyperacute ischemia [6][7][8][9]. Additionally, a positron emission tomography study demonstrated that oxygen extraction fraction (OEF) is reduced in acute IPH as opposed to ischemia, where OEF is increased [18].
Our results and the work of Stösser et al. further support this hypothesis: we were not able to show a difference in NIHSS at admission and discharge from hospital in patients with and without PDR, which would be expected if the PDR represents a frank ischemic region. Additionally, the finding that cases with higher PDR volumes showed significantly higher NIHSS scores at admission and discharge and had an increased mortality seems to be more attributable to higher initial IPH volumes. However, in contrast to our results, Orakcioglu et al. did not find perihemorrhagic ADC decreases in hyperacute IPH in a rat model [19].
Quantitative ADC values of the PDR were not studied before. We showed a continuous increase of absolute ADC values and rADC for up to 3 hours after SO (when excluding "wakeup"IPH cases and cases imaged >3 hours from SO due to low patient numbers). Therefore, diffusion restriction within the PDR is most pronounced directly after IPH onset and seems to diminish with time. This might be an explanation why other earlier reports studying the acute and subacute IPH phase (SO > 6 hours) did not describe a PDR. The PDR might primarily be related to a number of inflammatory actions and excitotoxic reactions caused by the blood clot [20]. Whether the overall decrease of PDR-ADC values within >3-6 hours from SO is due to the low patient number or not should be evaluated in further studies.
There are some limitations to our study. First, we may have missed cases that were suspected of hemorrhagic stroke (e. g. patients who did not show an abrupt SO or patients whose symptoms had not reached maximum intensity at SO) by MRI, because in our institution, these patients receive a computed tomography as initial imaging modality at admission. However, since MRI was routinely performed in patients with suspected stroke in a 6-hour time window, we do not assume a greater selection bias here. Secondly, due to the retrospective character of the study, sequence parameters differed by scanner and ADC values may not be directly comparable. Even though there is no common DWI quality assurance protocol, a study of Belli et al. showed a good agreement of a phantoms mean ADC value in a multicenter comparison [21]. In this study, we used relative ADC values/ratios that helped eliminating differences in the equipment. We did not assume systematic errors in the segmentation of partly small PDR volumes since we had 3 readers and MRI was shown to be very accurate in determining even minor tumor volumes [22]. Furthermore, perfusion imaging was not routinely performed after diagnosis of IPH in MRI; we were therefore not able to study perfusion changes in the PDR. Additionally, our patients did not routinely undergo a follow-up MRI (since CT is sufficient) that would have helped to further determine the FLAIR characteristics of the PDR in the acute or subacute period. Third, the number of patients imaged between 3 and 6 hours was too low to draw final conclusions about the PDR composition in this phase. Fourth, depending on the underlying sequence, an IPH typically shows different volumes (for example, SWI tends to overestimate IPH sizes, called "blooming effect") [16]. However, we accounted for this problem by spatial registration of SWI and ADC images after the predefinition of IPH ROI on SWI and subsequent careful manual correction of the ROIs, if applicable.
In conclusion, this is the first study precisely examining the perihematomal region in patients with hyperacute IPH by means of DWI/ADC mapping. We found a PDR adjacent to the perihematomal edema without corresponding FLAIR hyperintensities in a majority of patients. The degree of diffusion restriction was most pronounced directly after IPH onset and significantly declined in the following hours. Whether this phenomenon is caused by reversible hypoperfusion and/or metabolic changes has to be further studied to better understand pathophysiological processes involved in early IPH formation. Nevertheless, occurrence of PDR is not associated with worse clinical status and is therefore not a predictive marker for clinical status or short-term outcome.
Supporting information S1