Time course of the apparent diffusion coefficient (ADC) abnormality in human stroke

Diffusion-weighted MRI, which is based on the translational movement or diffusion of water,^1-3 can rapidly detect and localize ischemic brain injury as a hyperintense region.^4-8 Such hyperintense changes reflect in the acute stage a decline of the apparent diffusion coefficient (ADC) of water in ischemic brain tissue,^2,6 which is presumed to be related to intracellular water accumulation (cytotoxic edema),^4-6 decreased extracellular space,⁹ or changes in membrane permeability¹⁰ caused by rapid failure of energy metabolism. During the first minutes of ischemia-induced diffusion decrease, there is an associated decrease in the activity of the energy needing Na+ -K+-ATPase that maintains ionic gradients.¹¹ A decrease in ADC, similar to that observed in experimental ischemia, occurs with intraparenchymal injections of ouabain or excitotoxic agents.¹² The ADC decrease occurs with reductions in cerebral blood flow that cause dysfunction of Na+ -K+ -AT-Pase and disruption of ionic homeostasis.^13,14 This abnormality in diffusion-weighted MRI and ADC is observable within minutes after onset of experimental ischemia^{4,5,11,13,15-19} and occurs in human stroke before changes become apparent by 6 to 12 hours on conventional T₂- and proton density-weighted MRI.^7,8 The ADC becomes elevated in experimental stroke models and in human stroke in the chronic stage, which may reflect increased extracellular water content caused by vasogenic edema and cell lysis.^8,17,20-23

Several groups have demonstrated the feasibility of diffusion-weighted MRI for the acute diagnosis of stroke in humans.^7,8,22-24 MR diffusion and perfusion imaging in combination with conventional T₂-weighted MR images (T2WI) may have an important role in the acute assessment of stroke patients, because it allows the detection and monitoring of hyperacute ischemic lesions and their changes. Furthermore, it might permit the definition of irreversible or potentially salvageable, reversible ischemic damage.^{8,16,22,25,26} The potential role in monitoring therapeutic effects using these MRI methods increases the importance of defining the natural time course of these parameters in human stroke patients. Several studies^7,8,23,24 reported an ADC reduction in the acute stroke stage and a relative increase in ADC in the chronic stage. Although in experimental stroke models,^4,5,13,15-18 the hyperacute decrease in the ADC changed to a pseudonormalization or an elevation compared with normal brain tissue, the exact time boundaries of these changes have not been determined for human stroke. Pseudonormalization of the initial ADC reduction and the subsequent increase of ADC probably indicates a loss of cell membrane integrity(allowing less restricted movement of water molecules) and tissue necrosis.^17,21,22

The duration of the initial ADC reduction is debated because Warach et al.^7,8,27 found a prolonged ADC reduction, whereas Chien et al.²⁰ and Welch et al.^22,28 reported a more rapid change toward normalization and elevation within 24 hours from stroke onset. Therefore, we investigated whether ADC changes in acute human strokes are persistently low in a large sample of patients, and we sought to describe the temporal course of ADC evolution in human stroke.

Methods. Patients. We analyzed 157 diffusion MRI studies from 101 patients (61 men and 40 women), ranging in age from 22 to 94 years (mean age, 71 years) in which initial diffusion-weighted MRI showed a region of abnormal diffusion characteristic of ischemic stroke. All patients had an acute onset of signs and symptoms of cerebral ischemia. Patients with primary cerebral hemorrhage or other nonischemic diagnoses (e.g., abscess, brain tumor) were excluded. Time of onset from stroke was rounded to the nearest hour. Onset time was taken from the time the patient was last known to be without the new deficits. If a patient awoke with a new deficit, the time this patient went to bed or was last seen without this new deficit was taken as stroke onset time. The inclusion of patients who woke up with a deficit may introduce an error in defining the stroke onset time, which would be at maximum between 8 to 12 hours for a subset of patients studied.

Forty-five patients were studied more than once (usually two to three times; one patient was studied 6 times). Patients may have been treated with anticoagulant or antiplatelet drugs, but inclusion or analysis based on treatment was beyond the scope of this study. Patients who were studied multiple times had sustained neurologic deficits. No patients who had been treated with putative neuroprotective drugs were studied. All patients or authorized representatives gave written informed consent for the MR imaging procedure, which was approved by the Committee on Clinical Investigations of the Beth Israel Hospital.

Imaging Parameters. Most of the MRI studies (n = 89) were performed using a prototype whole-body 1.5 Tesla echo planar imaging (EPI) capable system (Siemens Medical Systems, Erlangen, Germany). The remaining studies (n = 68) were performed using a Siemens Vision 1.5 Tesla EPI system. A stimulation-optimized gradient coil set and a conventional gradient power amplifier in series with a high-voltage series resonant circuit, tuned at 1,000 Hz, was used. This achieved a maximum gradient amplitude of 35 mT/m with 250 µsec rise times for all the gradient axes. A circularly polarized head coil was used for excitation and signal reception. In addition to EPI diffusion-weighted MRI, conventional spin-echo T₂-weighted images (T2WI) and T₁-weighted images (T1WI) were acquired. Some patients were included in a sample that was previously reported.⁸

On the prototype whole-body 1.5-T EPI system, diffusion imaging was performed using a multi-slice, single-shot, spin-echo EPI sequence. Typical sequence parameters were TE (echo time) = 100 ms, matrix size 128 × 128, field of view (FOV) 250 millimeters, 7-millimeter slice thickness, no gap between slices. The diffusion gradient was applied in the transverse direction (x-direction). Diffusion gradient strength was varied between 0 and 30 mT/m by increments of 5 mT/m, resulting in seven b values ranging from 0 to 1271 sec/mm². Total acquisition time for all seven diffusion sensitivities was 48 seconds. For technical reasons, some of the measurements had results from fewer than seven b values.

On the Vision 1.5 Tesla EPI system, we used two b values (0 and 1000 sec/mm²). Typical imaging parameters were: TE = 118 milliseconds, matrix size 128 × 128, FOV 260 millimeters, 7-millimeter slice thickness, no gap between slices. The MR diffusion sequence at b = 1,000 was run three times; diffusion gradients were applied in successive scans in each of the x, y, and z directions. To minimize the effects of diffusion anisotropy, an average of the three diffusion directions was calculated to give the trace of the diffusion tensor. No special head restraints were used except the standard padding. No cardiac or respiratory gating was used. In all cases, the diffusion weighted sequence that was acquired with b = 0 sec/mm² was used as the T2WI.

Data analysis. Analysis of the diffusion changes was performed by calculating the ADC based on the Stejskal and Tanner equation¹ as the negative slope of the linear regression line best fitting the points for b versus 1n(SI); where SI is the signal intensity from a region of interest of the images acquired at each b value. Maps of ADC were created by performing this calculation on a pixel-by-pixel basis. To control for individual differences in ADC because of variables that could affect ADC values on a global basis, such as brain pulsation, temperature, or serum sodium concentration, ADC changes in the ischemic region were evaluated as rADC: mean ADC in a region of interest (ROI) in the center of the ischemic lesion was expressed as percent of a contralateral control region. These ROIs were drawn on a single axial slice in the core of the ischemic lesion; outer margin was approximately two to three pixels away from the rim of the diffusion-weighted MRI abnormality. In all initial diffusion-weighted MRI studies, ROIs were drawn on the image with the highest b value. At subacute to chronic time points the lesion was clearly evident in T2WI and less intense in diffusion-weighted MRI; therefore the ROI was chosen from the images that showed the fullest extent of the lesion. The ROI drawn on the diffusion-weighted MRI or T2WI was then transferred to the synthetic ADC map, and a mean ADC value within this region was obtained. For patients who showed a hemorrhagic transformation in follow-up scans, we avoided measuring the ADC in this area, because the presence of hemorrhage causes a signal loss on EPI images because of magnetic susceptibility effects and renders the ADC calculation meaningless. If this was not possible, the patient was excluded from the analysis. A nonischemic contralateral control ROI was carefully placed to exclude sulcal cerebrospinal fluid (CSF) which would have had high ADC values.

Statistical analysis. Data were analyzed by a one-way analysis of variance (ANOVA) using a univariate model with time from stroke onset as the grouping factor and rADC as the dependent measure. Mean rADCs for individual groups (table) were compared by a one-sample t test with the null hypothesis of rADC = 1. Significance was interpreted as p < 0.05 after Bonferroni correction for multiple t tests. Descriptive statistics were calculated for each group using values from each subject only once per group. Interobserver (G.S. and A.B.) reliabilities for rADC in a randomly selected subset (n = 40) were r = 0.98 (Pearson correlation). The mean ± SD of the interobserver differences for these 40 cases was 1.0 ± 6.5%.

Results. All patients in this study showed areas of hyperintensities on diffusion-weighted MRI and concomitant-reduced ADCs relative to the contralateral normal brain in their initial MRI studies. Of the 20 patients who were scanned within 6 hours of stroke onset, T2WI revealed abnormalities in only 1 of those 20 patients. In these hyperacute studies, signal abnormalities were most apparent with maximal diffusion sensitivity (usually b values greater than 900). No significant difference was seen between ADC values derived from diffusion-weighted MRI studies performed on the prototype 1.5 T EPI system or the Vision 1.5 T EPI system currently in use (p > 0.1 for all 10-time groups).

Mean ADC (SEM) for control regions in the 101 patients was 0.907 (0.090)× 10^-3 mm²/sec. Mean ADC (SEM) for studies performed at 3 hours or less was 0.581 (± 0.184) × 10^-3 mm²/sec, for lesions between 4 and 6 hours was 0.495 (± 0.143) × 10^-3 mm²/sec, for lesions between 7 to 12 hours was 0.465 (± 0.091) × 10^-3 mm²/sec, for 13 to 24 hours was 0.516 (± 0.086) × 10^-3 mm²/sec, for 25 to 48 hours was 0.528 (± 0.035) × 10^-3 mm²/sec, and for 49 to 96 hours was 0.638 (± 0.071) × 10^-3 mm²/sec. Chronic infarcts (>30 days) had an ADC value of 1.472 (± 0.154) × 10^-3 mm²/sec. Analysis of variance with time as the between-group factor revealed an overall effect of time (p < 0.001, F (9, 148) = 35.27). Group mean rADC values (in%) in the earliest 6 groups (until 96 hours after onset) and in the chronic group (>30 days after stroke) significantly (p < 0.005) deviated from 100% (ADC lesion = ADC control). rADC was significantly decreased in acute lesions (at least up to 96 hours) and tended to slightly increase during that time period (table). Diffusion-weighted MRI studies performed between 97 and 144 hours (5 and 6 days) after stroke still showed a reduced rADC (p < 0.03), but the deviation from 100% (ADC lesion = ADC control) was not significant after Bonferroni correction for multiple comparisons. rADC values from the 7th day to the 30th day after stroke onset were either slightly reduced, pseudonormalized, or elevated, and group differences were not significantly different from 100% (groups 8 and 9 in thetable).

Figure 1 demonstrates the time course of ADC abnormalities using all studies (triangles), including several time points for some patients. The degree of relative reduction varied somewhat between patients. All patients who were studied within the first 24 hours showed a reduction of at least 11%. The distribution of rADC values displayed two phases; the initial phase lasted for approximately 144 hours with a reduced rADC, and the late subacute to chronic phase showed a mean increase in rADC values but with greater variability. rADC values in between those two phases showed a heterogeneity with reduced, pseudonormalized, and elevated values. An example of diffusion-weighted MRI, ADC, and T2WI progression at six time points is illustrated for one patient infigure 2.

Discussion. This study provides evidence that the reduction in ADC observed in human stroke persists after stroke onset up to 6 days on average with a significant reduction for at least 96 hours. Analysis revealed two phases in the ADC time course: an early reduction and a late elevation. The first phase may last up to 144 hours (significant reduction only up to 96 hours) and is characterized by a relative ADC reduction compared with a contralateral control region. The second phase at late subacute to chronic time points is characterized by an increase in the mean rADC value. Variability was present in rADC values in this phase that trended from continued reduction initially to pseudonormalization and to elevation at later time points. These results agree with other studies with smaller patient samples and fewer imaging time points after stroke onset.^7,8,23,24 Although most of these studies agree that the hyperacute decline in the ADC persists for several days, few data points describe the change-over from reduced ADC values to its rise above normal.^23,29 The uncertain time of onset in patients who awoke with a deficit would not change the overall results of a prolonged reduction of rADC values for 4 days, because this error would be at maximum only 8 to 12 hours for a subset of patients in the study. Slight differences in the rADC values (particularly at chronic time points) between the current study and a previous study in our lab may be explained using different techniques. The turboSTEAM technique³⁰ was a preliminary approach and is not ideal for evaluating cerebral diffusion in acutely ill stroke patients. This approach was limited to single-slice acquisitions, had suboptimal diffusion sensitivity, and could have been contaminated by patient head motion or brain pulsation. Diffusion studies with the ultra-fast method of EPI³¹ can provide diffusion-weighted MRI of higher temporal resolution, effectively eliminating motion artifacts, and can permit whole brain studies in several seconds. Despite these differences, both studies revealed an initial phase with an rADC reduction and a later phase with an rADC pseudonormalization or elevation.

The approximate 41.7% decrease in ADC within the first few days after stroke onset agrees with results from animal studies that show a decrease in ADC values between 30-60% of normal values.^4,5,13,15-19 However, the time of changes to pseudonormalization or elevation of the ADC values differs between the animal and human studies, suggesting pathophysiologic differences between experimental ischemia models and human stroke. Variability in lesion rADC occurred thereafter, reflecting the variability in the evolution of human stroke. The prolonged decrease in rADC suggests that vasogenic edema and irreversible cell injury and death, expected to occur by several days, may not cause an immediate elevation of rADC.

Diffusion imaging is based on the modulation of signal intensity caused by water diffusion, and ADC values are the quantitative expression of the abnormal water diffusivity in brain parenchyma. Brain ADC changes are a function of intracellular-extracellular water homeostasis, and therefore are a sensitive marker of ionic equilibrium. Because disturbance of ion and water homeostasis and the subsequent excessive intracellular accumulation of sodium and water are among the first pathologic alterations induced by brain ischemia,^11,32,33 diffusion imaging is able to detect the injury within minutes. Measurements of the ADC can be used as an index of lesion severity and may also be useful in distinguishing reversible from irreversible ischemic injury.^{15,16,19,34-36} Although it is uncertain whether there is an absolute threshold of ADC reduction that predicts irreversible injury, there is compelling evidence from experimental ischemia models that ADC normalization within minutes to hours is a reliable indicator of cerebral recovery and correlates closely with recovery of energy metabolism.^19,34,37

Results from our laboratory and recent findings in other laboratories contrast with reports by Chien et al.²⁰ and Welch et al.²² who suggested that normalization and elevation of ADC might occur by 24 hours after stroke onset. These differences in the ADC time course could be explained by technical factors related to different techniques of measurement and analysis.²⁷ In our 106 measurements up to 48 hours from stroke onset, none of the rADCs was elevated; of 135 measurements in 90 patients up to 144 hours from onset only 3 measurements were greater than 100% (119% at 60 hours, 103% at 100 hours, and 118% at 144 hours). In one of the earliest reports concerning diffusion-weighted MRI in human stroke, Chien et al.²⁰ reported elevated ADCs in most stroke patients who were studied, although two of the three patients that were studied within 24 hours after onset had ADC values within the range of a normal control group of younger volunteers; no lesions had reduced ADC. The younger sample of control subjects in that study were likely to have lower baseline ADC values,³⁸ and thus, reduced relative ADC in the older patient sample was more likely to appear normal or elevated in comparison with the younger control group. We cannot rule out that other factors, such as differences in the underlying stroke etiology (e.g., thrombosis versus artery-to-artery embolism) or early reperfusion because of a sufficient collateral circulation, could have accounted for the earlier ADC normalization reported in other studies.^22,28

There may be heterogeneity of ADC values within acute ischemic lesions. However, because we chose to characterize the ischemic lesion as a whole, this was not addressed in the current study. Furthermore, the outer rim of the lesion may have ADC abnormalities of unique pathophysiologic properties in which ATP content may be preserved at the earliest time points, despite ADC decreases because cortical spreading depression may be present.^39-42 Our method of ROI analysis deliberately excluded the outer rim (approximately 4 to 6 millimeters in plane) as a precaution against partial volume effects causing inaccuracies in the measurements. The present study did not address the issue of probable intralesion ADC heterogeneity or the ADC at the rim of the lesion. Intralesion ADC heterogeneity is better examined using a method such as that used by Welch et al.,²² whereas ADC heterogeneity between the lesion center and rim is best addressed with high-resolution(submillimeter) MRI techniques.

After the hyperacute period, ADC may reflect a mixture of pathologic events, including persistence or evolution of acute cellular injury, the development of so-called vasogenic edema, and evolution of ischemic cell death, cell lysis, and membrane disruption. The latter two events would preferentially increase extracellular volume and would allow less restricted movement of water molecules, and thereby increase the ADC, leading to pseudonormalization or elevation.²¹ We have previously noted that no MRI parameter can reliably distinguish tissue in the hyperacute, acute, subacute, or chronic time periods.⁸ At the hyperacute period (0 to 6 hours), diffusion-weighted MRI signal intensity is elevated and ADC is decreased, whereas T2WI signal intensity is typically normal, indicating no significant effect of vasogenic edema. At subacute or chronic time points, the diffusion-weighted MRI signal intensity value may be low, normal, or elevated and is not sufficient to determine the direction of the ADC change. The ADC value alone, or the T2WI signal intensity alone, are also insufficient to determine the stage of infarct evolution; multiple parameters are necessary. Combining ADC and T2WI to correlate MR abnormalities with stages of tissue histopathology has been a promising approach suggested by Knight et al.¹⁷ and Welch et al.²² It may permit characterization of stages of ischemic lesion evolution and may lead to a method of distinguishing reversible from irreversible ischemic injury based on pathophysiology rather than time windows. The combination of MR perfusion imaging with diffusion imaging^{5,24,37,41,42,43} will most probably provide an important additional marker for characterizing tissue salvageability. This is particularly true in the first 12 hours after stroke onset when predicting and identification of potentially salvageable tissue is of greatest interest. The ADC is a noninvasive, quantitative parameter of water diffusivity in brain parenchyma that is sensitive to the earliest pathophysiologic changes in hyperacute ischemic stroke. Defining the natural time course of ADC changes in the evolution of human stroke may provide important information for the distinction between salvageable and nonsalvageable tissue and may be useful for monitoring therapeutic interventions.

Acknowledgments

We thank Venkatesan Thangaraj for assistance in data analysis and Drs. Alison Baird and Karl Lovblad for helpful comments.