Voxel-based analysis of apparent diffusion coefficient in perihaematomal oedema: associated factors and outcome predictive value for intracerebral haemorrhage.

Objectives The pathophysiology of perihaematomal oedema (PO) surrounding a primary intracerebral haemorrhage (ICH) is complicated and incompletely understood. We prospectively investigated the components of PO with voxel-based analysis of the apparent diffusion coefficient (ADC) value and assessed its predictive value for functional outcome. Design Forty-six patients with ICH who were enrolled for clinical evaluation underwent MRI scans within 24 h after ICH. Based on the ADC value of the ipsilateral voxels divided by the mean ADC value of the contralateral mirror region of interest, the voxels with oedema were classified into three categories: cytotoxic, vasogenic and undetermined. The percentages of cytotoxic and vasogenic oedema were then calculated and correlated with clinical outcome according to the modified Rankin Scale (mRS) at 6 months after ICH. The intraobserver and interobserver reliability of this method were examined using intraclass correlation coefficients. Results The intraclass correlation coefficients showed that analysis using the voxel-based method is highly reliable. Among the clinical variables tested, age and serum creatinine levels were positively correlated with percentage of cytotoxic oedema. Age, history of coronary artery disease, National Institutes of Health Stroke Scale score and percentage of cytotoxic oedema were all associated with mRS at 6 months after ICH. Conclusions The pathophysiological processes within PO are complicated. Voxel-based analysis of ADC values may help to identify the components of PO and may be beneficial for decision making and predicting outcome.

Introduction

Perihaematomal oedema (PO) develops within the first few days after primary intracerebral haemorrhage (ICH). Whether or not PO contributes to ICH-induced neurological deficits and patient outcome is still controversial and deserves further investigation.1 2 The pathophysiological processes within PO are complicated and may provide valuable clues regarding outcome.2 Diffusion MRI, a technique that can be used to probe tissue microstructures by measuring the molecular diffusion of water, can help characterise the components of oedema. Cytotoxic oedema decreases the apparent diffusion coefficient (ADC), whereas vasogenic oedema increases the ADC.3 4 Diffusion MRI has been applied to the study of perihaematomal injury in patients with ICH but with inconsistent results.5–10 In this study, we investigated the components of PO in 46 consecutive patients with ICH using voxel-based analysis of the ADC value.

Methods

Patients

Approval for the study was obtained from the Institutional Review Board of Chang Gung Memorial Hospital, and informed consent was obtained from the patients or their relatives. Forty-six consecutive patients with a first ICH and with no history of any neurological deficits were enrolled. Patients with any contraindication to undergoing MRI, requiring emergent surgery, with a history or imaging findings of previous ICH or other neurological insult, or with evidence of intraventricular haemorrhage or haemorrhage related to a tumour, trauma, coagulopathy or vascular lesion were excluded.

The following clinical data were recorded within 24 h after ICH: age, sex, blood pressure, blood glucose level, haemoglobin level, white blood cell count, platelet count and creatinine level. The National Institutes of Health Stroke Scale (NIHSS) and Glasgow Coma Scale scores were estimated by two experienced neurosurgeons within 24 h after ICH. The Modified Rankin Scale (mRS) score was estimated by the same neurosurgeons at 6 months after ICH and was defined as patient outcome.

MRI

MRI scans were obtained with a 1.5 T human MRI scanner (Gyroscan Intera; Philips Medical Systems, Best, The Netherlands) within 24 h after symptom onset. Standard sequences included axial T2*-weighted gradient echo images for the location and volume of the haematoma and axial fluid-attenuated inversion recovery (FLAIR) images for PO extension. The area of haematoma was manually segmented on T2*-weighted gradient echo images and the area of oedema was manually segmented on FLAIR images. The volumes of oedema and haematoma were then estimated by the ImageJ processing program (http://rsbweb.nih.gov/ij/). Diffusion-weighted images were acquired using a diffusion-sensitised echo-planar imaging pulse sequence. Diffusion sensitivity (b1=1000 s/mm2) was applied sequentially in the x, y and z gradient directions, and a reference image without diffusion sensitivity (b0≈0 s/mm2) was acquired. The ADC map was derived directly from these diffusion-weighted images.

Voxel-based analysis of ADC

Voxel-based analysis of ADC values was performed using ‘in-house’ software developed at MATLAB (Math-Work, Natick, Massachusetts, USA). The analytical process is illustrated in the figure 1.

Figure 1

Flow diagram for voxel-based analysis of the apparent diffusion coefficient (ADC). (A) The region of interest, perihematomal edema (PO), was outlined on an ADC map by an experienced neuroradiologist. The PO was marked following inspection of all available imaging data. (B) The mean ADC value of the contralateral mirror region of interest (ROI) was calculated. The relative ADC for each voxel in PO was defined as the ADC value of the ipsilateral voxel divided by the mean ADC value of the contralateral mirror region of interest. (C) Generation of a relative ADC histogram, where the x-axis represents the relative ADC value and the y-axis is scaled with the number of voxels at any relative ADC value. (D) Voxels within PO were stratified into three categories based on the relative ADC value and the percentage of each category was then calculated.

PO was manually segmented as the lesion-side region of interest (ROI) on the MRI images by an experienced neuroradiologist. The PO was marked following inspection of all available imaging data including T2-weighted images, fluid-attenuated inversion recovery images and ADC maps. The equation for the straight line of the brain midline was then defined and calculated. The mirror ROI on the contralateral normal hemisphere across the brain midline was then defined automatically according to the coordinates of the lesion-side ROI and the straight line equation. The relative ADC value for each voxel in PO was calculated as the ADC value of the voxel in the lesion-side ROI divided by the mean ADC value of the mirror ROI. The histogram was then calculated to represent the distribution of the relative ADC value. Voxels within PO were stratified into three categories based on the relative ADC value and the percentage of each category was then calculated.

Statistical methods

The values for baseline characteristics were presented as means and SD. Univariate and multiple stepwise linear regression models were used to analyse the correlation between clinical variables and percentages of cytotoxic and vasogenic oedema. Univariate and multiple stepwise linear regression models were used to analyse the relationship of possible predictor variables to continuous clinical outcome (mRS). p Values less than or equal to 0.05 were deemed significant.

To validate the reliability of the voxel-based analytical method in giving the same result on different occasions (intraobserver reliability) or between different neuroradiologists (interobserver reliability), we examined the percentage of cytotoxic oedema measured by a neurologist (Dr Tsai) on different occasions and by two neuroradiologists (Dr Weng and Dr Tsai) with intraclass correlation coefficients (ICC). A one-way ICC with absolute agreement was used to assess intraobserver reliability and a two-way ICC with absolute agreement was used to examine interobserver reliability. All statistical analyses were performed using Stata V.11.0 statistical software (StataCorp).

Results

The baseline characteristics, and clinical and radiological features of the 46 enrolled patients are presented in table 1.

Table 1

Baseline characteristics of 46 patients with intracerebral hemorrhage

Age (years)	65.2±12.7
Male (%)	25 (54.3)
Location of haematoma (%)
Thalamus	17 (37.0)
Basal ganglia	12 (26.1)
Putamen	11 (23.9)
Lobar	5 (10.9)
Cerebellum	1 (2.1)
Medical history (%)
Hypertension	34 (73.9)
Antihypertensive medication	20 (43.5)
Diabetes	9 (19.6)
Coronary artery disease	3 (6.5)
Smoker	12 (26.1)
Alcoholism	11 (23.9)
Blood pressure (mm Hg)
Systolic	191.3±22.5
Diastolic	106.3±16.3
Mean arterial	133.6±15.4
Serum glucose (mmol/l)	143.4±59.8
Haemoglobin (g/dl)	14.1±1.5
Platelet count (1000/μl)	193.7±118.1
Creatinine (mg/dl)	1.0±0.4
White blood cell count (1000/μl)	8.1±2.5
NIHSS score within 24 h	10.6±5.3
GCS score within 24 h	13.3±2.6
mRS score at 6 months	2.9±1.7
Haematoma volume (ml)	19.4±14.2
Oedema volume (ml)	14.3±14.2
Percentage of vasogenic oedema (%)	55.7±25.2
Percentage of cytotoxic oedema (%)	15.9±16.7

GCS, Glasgow Coma Scale; mRS, Modified Rankin Scale; NIHSS, National Institutes of Health Stroke Scale.

The ICC showed high reliability for measuring the percentage of cytotoxic oedema (ICC=0.978 for intraobserver reliability; ICC=0.933 for interobserver reliability).

Table 2 shows the correlation coefficients of clinical, laboratory and radiological variables in relation to percentages of cytotoxic oedema. As indicated, there were significant positive correlations between both patient age and creatinine level, and percentage of cytotoxic oedema (p=0.003 and p=0.021, respectively). There was also a negative correlation between haemoglobin level and percentage of cytotoxic oedema (p=0.005). In multivariate analysis including variables that were positively associated with the percentage of cytotoxic oedema in univariate analysis, age and creatinine level remained significantly associated with percentage of cytotoxic oedema.

Table 2

Correlation between clinical variables and percentage of cytotoxic oedema

	Univariate analysis		Multivariate analysis
	Coefficient (95% CI)	p Value	Coefficient (95% CI)	p Value
Age	0.572 (0.212 to 0.931)	0.003*	0.534 (0.190 to 0.879)	0.003*
Location of haematoma	−0.904 (−5.444 to 3.636)	0.690
Medical history (%)
Hypertension	1.289 (−10.126 to 12.704)	0.821
Diabetes	1.470 (−11.165 to 14.105)	0.816
Coronary artery disease	−4.052 (−24.878 to 16.774)	0.697
Smoker	−7.314 (−21.730 to 7.101)	0.312
Alcoholism	5.940 (−8.880 to 20.759)	0.423
Haemoglobin level	−4.595 (−7.764 to −1.426)	0.005*
Creatinine level	13.320 (2.100 to 24.540)	0.021*	11.822 (1.526 to 22.119)	0.025*
White blood cell count	1.394 (−0.555 to 3.3444)	0.157
Systolic BP	0.113 (−0.110 to 0.336)	0.314
Diastolic BP	−0.018 (−0.329 to 0.293)	0.908
Mean arterial BP	−0.061 (−0.565 to 0.443)	0.804
Platelet count	−0.008 (−0.051 to 0.035)	0.703
Serum glucose	−0.018 (−0.103 to 0.066)	0.667
GCS score	−1.410 (−3.342 to 0.523)	0.149
NIHSS score	0.788 (−0.136 to 1.713)	0.093
Haematoma size	0.000 (0.000 to 0.000)	0.555
Oedema volume	0.000 (−0.000 to 0.001)	0.374

Significant differences were defined as those with p<0.05.

BP, blood pressure; GCS, Glasgow Coma Scale; NIHSS, National Institutes of Health Stroke Scale.

Table 3 shows the correlation coefficients of clinical, laboratory and radiological variables in relation to percentages of vasogenic oedema. There was a significant positive correlation between patient age and percentage of vasogenic oedema (p=0.000) and a negative correlation between haemoglobin level and percentage of vasogenic oedema (p=0.017). In multivariate analysis including variables that were positively associated with the percentage of vasogenic oedema in univariate analysis, age remained significantly associated with the percentage of vasogenic oedema.

Table 3

Correlation between clinical variables and percentage of vasogenic oedema

	Univariate analysis		Multivariate analysis
	Coefficient (95% CI)	p Value	Coefficient (95% CI)	p Value
Age	−0.990 (−1.513 to −0.466)	0.000*	−0.990 (−1.513 to −0.466)	0.000*
Location of haematoma	2.155 (−4.692 to 9.001)	0.529
Medical history (%)
Hypertension	−3.754 (−20.987 to 13.479)	0.663
Diabetes	2.176 (−16.929 to 21.282)	0.819
Coronary artery disease	8.069 (−22.548 to 38.686)	0.598
Smoker	10.359 (−6.622 to 27.341)	0.225
Alcoholism	−1.713 (−19.485 to 16.068)	0.847
Haemoglobin level	6.031 (1.126 to 10.935)	0.017*
Creatinine level	−6.386 (−24.317 to 11.546)	0.477
White blood cell count	−1.518 (−4.500 to 1.463)	0.310
Systolic BP	−0.076 (−0.416 to 0.265)	0.657
Diastolic BP	0.304 (−0.156 to 0.765)	0.190
Mean arterial BP	0.237 (−0.464 to 0.937)	0.492
Platelet count	0.022 (−0.043 to 0.086)	0.502
Serum glucose	0.080 (−0.046 to 0.206)	0.206
GCS score	1.028 (−1.949 to 4.004)	0.490
NIHSS score	−0.656 (−2.086 to 0.775)	0.361
Haematoma size	0.000 (0.000 to 0.001)	0.343
Oedema volume	0.000 (0.000 to 0.001)	0.691

Significant differences were defined as those with p<0.05.

BP, blood pressure; GCS, Glasgow Coma Scale; NIHSS, National Institutes of Health Stroke Scale.

Table 4 shows the predictors of functional outcome at 6 months after ICH. The clinical, laboratory and radiological variables that were positively associated with mRS at 6 months in univariate analysis were age (p=0.000), history of coronary artery disease (p=0.025), NIHSS score (p=0.003) and percentage of cytotoxic oedema (p=0.001). Haemoglobin level (p=0.010) and percentage of vasogenic oedema (p=0.041) were negatively correlated with mRS at 6 months. In multivariate analysis including variables that were positively associated with mRS at 6 months in univariate analysis, age, history of coronary artery disease, NIHSS score and percentage of cytotoxic oedema remained significantly associated with mRS at 6 months.

Table 4

Predictors of functional outcome at 6 months after intracerebral hemorrhage

	Univariate analysis		Multivariate analysis
	Coefficient (95% CI)	p Value	Coefficient (95% CI)	p Value
Age	0.065 (0.031 to 0.100)	0.000*	0.043 (0.013 to 0.074)	0.007*
Location of haematoma	−0.246 (−0.690 to 0.206)	0.206
Medical history (%)
Hypertension	0.475 (−0.655 to 1.606)	0.401
Diabetes	0.910 (−0.321 to 2.141)	0.143
Coronary artery disease	2.209 (0.297 to 4.122)	0.025*	2.339 (0.931 to 3.747)	0.002*
Smoker	−0.363 (−1.497 to 0.772)	0.523
Alcoholism	−0.751 (−1.902 to 0.400)	0.195
Haemoglobin level	−0.427 (−0.747 to −0.107)	0.010*
Creatinine level	0.856 (−0.305 to 2.017)	0.145
White blood cell count	0.133 (−0.062 to 0.328)	0.175
Systolic BP	−0.009 (−0.031 to 0.013)	0.424
Diastolic BP	−0.019 (−0.050 to 0.011)	0.205
Mean arterial BP	−0.027 (−0.074 to 0.019)	0.236
Platelet count	0.004 (−0.001 to 0.008)	0.096
Serum glucose	0.004 (−0.004 to 0.013)	0.270
GCS score	−0.126 (−0.320 to 0.068)	0.198
NIHSS score	0.135 (0.049 to 0.221)	0.003*	0.116 (0.048 to 0.184)	0.001*
Haematoma size	0.000 (−0.000 to 0.000)	0.566
Oedema volume	0.000 (−0.000 to 0.000)	0.232
Cytotoxic oedema	0.047 (0.021 to 0.074)	0.001*	0.026 (0.002 to 0.050)	0.037*
Vasogenic oedema	−0.020 (−0.039 to −0.001)	0.041*

Significant differences were defined as those with p≤0.05.

BP, blood pressure; GCS, Glasgow Coma Scale; mRS, Modified Rankin Scale; NIHSS, National Institutes of Health Stroke Scale.

Discussion

Diffusion MRI has been used to study perihaematomal injury in patients with ICH, but with inconsistent results.5–14 In general, the mean ADC values in the perihaematomal regions relative to contralateral, homologous brain regions may be elevated early during the acute stage with peak increases noted 2–3 days after ICH. Nevertheless, decreased relative mean ADC values can be observed in some patients and have been reported to be associated with poor clinical outcome.4 8 13 14 However, the overall mean ADC value within ROI, as used in most studies of PO, may be the result of reduced ADC due to cytotoxic oedema or ischaemia being cancelled out by the effect of vasogenic oedema which would elevate the ADC value, thus decreasing the sensitivity and specificity for predicting outcome.11 This may be the reason why the results of the different diffusion MRI studies were controversial and inconsistent.

The first aim of this study was to clarify the factors associated with cytotoxic oedema formation in the perihaematomal zone. Based on previous studies of ischaemic stroke, a 10% or greater reduction in ADC value is evidence of cytotoxic oedema and ischaemic injury.3 For ICH, the possible mechanisms of decreased perihaematomal ADC values are cytotoxic oedema and neuronal injury which may result from ICH mass effect, inflammation or toxin injury from blood breakdown products such as thrombin or iron.4 5 8 11 In our study, we found that the percentage of cytotoxic oedema is associated with patient age and creatinine level. Age has been reported to be an independent contributor to outcome after ICH14 and was positively correlated with the percentage of cytotoxic oedema. Older age may contribute to a weaker systemic response to acute ICH and result in more ischaemic and neuronal injury. Older rats with ICH were found to have severe brain swelling and greater perihaematomal induction of stress proteins but a weaker astrocytic reaction to haematoma.15 Serum creatinine level has been reported to be associated with haematoma growth.16 Creatinine is a recognised marker of long-standing hypertension with accumulated vascular injury and increased fragility of small vessels. This condition may lead to more ischaemic injury after ICH owing to poor vascular auto-regulation.

The second aim of our work was to identify the outcome prediction value of this voxel-based analytical method. We found that the percentage of cytotoxic oedema is positively correlated with mRS at 6 months after ICH. Decreased relative mean ADC values have been reported to be associated with poor clinical outcome.8 Cytotoxic oedema is due to the derangement in cellular metabolism which results in inadequate functioning of the sodium-potassium pump in the glial cell membrane. As a result, cell swelling, cell lysis, necrosis and irreversible cell death all occur. Vasogenic oedema, on the other hand, is due to penetration of intravascular proteins and fluid into the cerebral parenchymal extracellular space after a breakdown of tight endothelial junctions which make up the blood–brain barrier. Vasogenic oedema is often recognised as a reversible injury. The results of this study may help to clarify the debate on whether there is a perihaematomal ischaemic penumbra.4 5 11 17 Cytotoxic and vasogenic oedema coexist in the perihaematomal area. Perihaematomal regions with cytotoxic oedema may develop irreversible neural injury. The outcome for oedematous brain tissue and its effects upon patient prognosis depend not only on the mass effect but also on the underlying neural damage, such as cytotoxic and vasogenic oedema.

There are several limitations to this study. First, we excluded patients with extensive haemorrhage that required emergent surgical evacuation. Haematoma volumes in this study ranged between 1.5 and 65.9 ml (mean 19.4 ml; SD 14.2), which were relatively small, and may explain why haematoma and oedema size were not related to functional outcome in this study. Second, our results were derived from a small sample size and further group analysis of different haematoma locations could not be carried out. Third, the automatically selected mirror ROI may have included cerebrospinal fluid space. Therefore, ADC values calculated in this area could have been affected by methodological imprecision. Furthermore, the definite thresholds of cytotoxic and vasogenic oedema established by ADC have not yet been validated in a wide population of patients or in animal studies. Finally, because PO peaks between 10 and 20 days after ICH in humans,2 18 an additional MRI scan at this time point may have provided more information regarding diffusional and pathological changes within the oedema zone.

Conclusion

The pathophysiological processes within PO are complicated. Voxel-based analysis of ADC appears promising based on this study. It can help identify the components of PO and may be useful for decision-making and predicting outcome. Further research should be carried out to determine how the factors which contribute to PO affect the ADC and how this analytical method can be used for predicting the result of treatment and providing individualised therapy.