Anoxic Brain Injury Detection with the Normalized Diffusion to ASL Perfusion Ratio: Implications for Blood-Brain Barrier Injury and Permeability

DISCUSSION

Anoxic brain injury is a medical condition with devastating outcomes.^12,33,34 Discriminating anoxic from nonanoxic injuries, especially in the neonatal period, is important for medical management,³⁵ potential therapeutic hypothermia protocol adjustment,²² and prognostic estimation.^13,34,35 Classic findings of diffusion restriction as well as quantitative hyperperfusion, especially in the regions of higher metabolic demand such as the basal ganglia and cortex, have been described previously in the literature.^{1⇓-3,12⇓-14,22} However, there remains a gap in the literature regarding the anoxic brain injury patterns on nonquantitative ASL sequences. In this article, we characterized the qualitative ASL perfusion findings in anoxic injury. We identified the relationship between ASL perfusion and diffusion restriction, which demonstrated a homogeneously positive correlation in patients with anoxic brain injuries so that areas of restricted diffusion showed increased ASL perfusion signal (Fig 5D–F).

• Download figure
• Open in new tab
• Download powerpoint

Fig 5.

Quantitative-versus-nonquantitative ASL perfusion in anoxic brain injury. A–C, Quantitative perfusion imaging in a 31-year-old patient who had an anoxic brain injury after cardiac arrest. B = 1000 DWI (A) and ADC map (B) show diffusion restriction in the posterior putamen (arrows) and occipital cortex. C, Quantitative pulsed ASL image shows global hyperperfusion. Subtle increased cerebral blood flow is seen in the posterior putamen and occipital lobes, corresponding to the areas of diffusion restriction (arrows). D–F, Nonquantitative perfusion imaging in a 22-month-old child who had an anoxic injury after cardiac arrest. B = 1000 DWI (D) and ADC map (E) show diffusion restriction in the basal ganglia (arrowheads) and occipital cortex. F, Nonquantitative colorized pulsed ASL image demonstrates marked increases in the ASL signal in the putamen (arrowheads) and occipital cortex, corresponding to the areas of diffusion restriction.

Review of the 35 patients with anoxic brain injuries demonstrated classic diffusion restriction findings (hyperintense on DWI and hypointense on ADC). Ischemic changes including energy metabolism variations and cytotoxic edema reflected by diffusion-weighted imaging were prominent in the metabolically active tissue in the brain,³⁶ most commonly in the basal ganglia (especially the posterior putamina), ventrolateral thalami, and cortex.^{1⇓-3,37⇓-39} In addition to a pattern of global hyperperfusion seen with prior quantitative ASL studies (Fig 5A–C), review of the qualitative ASL perfusion images also demonstrated regional signal increases that correlated with the areas of increased DWI signal (Fig 5D–F).^12,13 We suspect that the regional ASL perfusion differences are detectable with quantitative techniques, but on typical color maps that might not be adequately scaled, global hyperperfusion may render these relative differences less apparent because the entire brain is at the high end of the color spectrum.

ROI analyses demonstrated a uniform homogeneously positive relationship between the ASL perfusion signal and the DWI signal in the selected regions among patients with anoxic brain injuries (Fig 2). This relationship held true not only in the more commonly described basal ganglia region but also in the cortical gray matter and white matter regions.^15,21 Analysis of the ADC signal revealed a similar relationship, consistent with prior literature demonstrating significant correlations between decreased ADC and elevated ASL cerebral blood perfusion signals;²⁵ however, the diagnostic performance of the ADC and normalized ADC was inferior, likely due to higher susceptibility to volume-averaging artifacts from the surrounding CSF or intrinsic properties yet to be delineated causing higher ASL cerebral blood perfusion elevation than the corresponding ADC abnormality.²⁵

This homogeneously positive relationship between ASL and DWI may be related to changes in global BBB permeability. Tanaka et al⁹ previously showed that the ASL perfusion sequence tends to overestimate CBF in ischemic stroke in an animal model. Unilateral injection of mannitol in healthy animals, which increased the permeability of the BBB, also increased the ASL signal. Assuming that the magnetically tagged water molecules are freely diffusible across the BBB,^5,10,11 they postulated that ischemic stroke injuries, with subsequent disruption of the BBB, allowed increased extravasation of water molecules, implying that ASL perfusion signal is dependent on BBB integrity.

On the basis of the results of the animal studies and our current observations, we hypothesize that in an anoxic injury, there is both a global and regional disruption of the BBB, especially in the regions of higher metabolic demands, leading to regional increases in ASL signal that correlate with the areas of greatest diffusion restriction (Fig 5D–F). Nondependence of the positive relationship between ASL and DWI signal on the time interval from injury to imaging (Fig 3) suggests that the BBB injury is present and enduring up to 14 days (range, 0–14 days after the reported inciting event). This BBB disruption is normally seen on gadolinium contrast-enhanced MR imaging 2–3 days after ischemic injury and may persist for months;⁴⁰ however, it is conceivable that the BBB injury occurs earlier than 2–3 days, and it is only after 2–3 days that the BBB disruption is large enough for the gadolinium particles to cross the BBB.

Gadoterate meglumine, a commonly used chelated gadolinium contrast medium, is 2 nm,⁴¹ compared with 0.275 nm for a water molecule. ASL uses the protons within the blood water,^4,5,42,43 making the ASL sequence much more sensitive to early changes in the BBB pore size. Although there is not a significantly nonzero slope, positive trends of both the linear regression slope and R² over interval to imaging were observed (Fig 3). These may suggest that the repair of the BBB is gradual and lags behind the pseudonormalization of the diffusion signals, a phenomenon in which diffusion-weighted imaging improves and appears to have normal findings by the end of the first week.³ Gradually decreasing metabolic demands due to delayed cell death during the energy failure component of the reperfusion phase^21,44,45 with persistent BBB disruption could theoretically have contributed to a similar relationship. Further research with animal models and survival studies is needed to delineate the underlying physiologic response and to evaluate the effects of BBB disruption in humans and the recovery of BBB integrity.

The animal studies showing an increase in ASL signal with BBB disruption alone highlight 1 potential limitation of quantitative ASL techniques. Quantitative ASL equations are based on the assumption that the ASL tag is a freely diffusible tracer and is able to freely cross the BBB. If the animal studies and our hypothesis are true, then nonquantitative methods may be advantageous when studying entities such as anoxic brain injury. Prior quantitative studies looking at anoxic brain injury showed marked global hyperperfusion with mean gray matter perfusion values reaching up to 204 mL/100 g of tissue/min (Fig 5A–C).¹² These exceptionally high perfusion values are 3–4 times normal and would mean between 45% and 60% of cardiac output would be directed to the brain.⁴⁶ On the basis of these superphysiologic values, we believe that the cerebral perfusion has increased to a degree; however, changes in BBB permeability have also influenced the quantitative ASL measurements. Future studies may evaluate quantitative and nonquantitative ASL perfusion in a human BBB disruption model to validate this hypothesis.

Another potential cause of the diffusion and perfusion ratio observations would be loss of autoregulation, leading to loss of global vasoreactivity after anoxic brain injury.^47,48 Loss of vasoreactivity has been suggested in neonates with hypoxic-ischemic encephalopathy.⁴⁹ Additionally, global hyperperfusion affecting the white matter could artificially increase the ratio in the white matter regions by increasing their perfusion values to a level similar to that of the gray matter. Future studies could investigate this possibility by applying gray-white matter segmentation and analyzing the mean diffusion-to-perfusion ratios in the gray matter and white matter, respectively.

Among healthy controls, significantly lower R² values and lower slopes of the linear regression models implied underlying heterogeneity of the relationship between the ASL perfusion and the DWI signal. Analyses of the ADC also demonstrated similar results. The lack of a homogeneous relationship is likely secondary to a much lower signal-to-noise ratio among the intracranial structures in healthy controls. Small variability in metabolic demand, regional perfusion, BBB permeability, and the inherent delay of the autoregulation feedback loop may also contribute to the demonstrated heterogeneities.^47,48

Generation of the voxelwise NDP color map highlights 2 key concepts. First, in patients with anoxic brain injuries, for reasons that remain elusive, the ratio between ASL and DWI signal was homogeneous throughout the brain, including the white and gray matter (Fig 4A, -B). In patients with anoxia, this voxelwise global homogeneous relationship supersedes any larger manual or automated ROI methodology by showing that as long as enough ROIs are selected and the locations of the DWI and ASL are matched, the actual location of the ROI in a particular structure is not significant. Second, in healthy controls, the NDP color map (Fig 4E, -F) demonstrated global heterogeneity, evident by the large distribution of DWI/ASL ratios through the brain structures. Thus, in the healthy population, ROI placement and, most important, the number of ROIs become significant due to the global heterogeneity. This is most pronounced in the white matter regions where perfusion values are very low. The NDP color map of the predominately unilateral case, which was an unfortunate patient with nonaccidental trauma with anoxic brain injury secondary to unilateral carotid occlusion from strangulation, demonstrated more uniform color distribution in the affected right hemisphere and heterogeneous color distribution in the nonaffected left hemisphere (Fig 4I–J). With future automatic implementation, we believe the NDP color map can be an additional imaging biomarker for anoxic brain injuries, especially in the neonatal population. The NDP ratio map avoids the dependence, complexity, and inherit limitations of manual or automated ROI selection and ratio calculations.

In an attempt to simplify the NDP ratio generation for a neuroradiologist in a general clinical practice who might not have a pipeline to generate these NDP maps or have time to select 10 ROIs, we examined the diagnostic performance of choosing 2 random ROIs. The positivity of the slope of the line between the 2 points on a graph of ASL-versus-DWI signal was determined. Positive slopes would be consistent with anoxic injuries. Negative slopes would be consistent with healthy patients. The diagnostic performance of this method was inferior to that of the 12-point ROI linear regression method. This was expected because the inherent heterogeneity of the NDP ratios in the healthy population could not be accurately captured by a random 2-point selection (Fig 4E, -F). If the NDP ratio color map cannot be obtained while one is trying to confirm the diagnosis of anoxic brain injury, alternatively, 2 different basal ganglia or thalamus structures from the opposite hemisphere can be selected for slope determination with reasonable sensitivity (0.78) and specificity (0.71).

This study has several limitations, in addition to the inherent limitations of being a retrospective study. First, despite recent progress in the understanding of BBB permeability and ASL perfusion, our knowledge of their characteristics, especially in the context of anoxic brain injuries, remains limited. The exact underlying physiologic property that the NDP ratio is reflecting remains uncertain. Further research is needed for better characterization of the physical and molecular properties of the BBB during anoxic brain injury, timing, and its subsequent repair and also to evaluate the NDP ratio in models of BBB disruption. Second, we performed a strict comparison between patients with a confirmed diagnosis of anoxic brain injury and control subjects with normal imaging examination findings. This improves our statistical power and highlights our findings. Further studies will aim to examine the NDP ratios in other isolated and mixed etiologies, including tumor, stroke, and metabolic diseases. Third, we used the in-line vendor software for automatic ADC map generation. Prior comparative studies have demonstrated significant variations in the generation of ADC maps across different postprocessing software.⁵⁰ Future studies could be repeated with an open-source postprocessing software, such as Horos (https://sourceforge.net/projects/horos/). Moreover, manually placed ROI regions could have signal interference from surrounding structures, volume averaging, or incompletely matched locations. An automated atlas-based process for brain parcellation, region selection, and location-matching across series could be more accurate if performed with a high-resolution T1 series.

Last, the ASL perfusion sequence remains suboptimal in the neonatal population due to the small head size and requires optimization to ensure good-quality imaging. Our postlabel delay times at 1200 ms differ from the International Society for Magnetic Resonance in Medicine Perfusion Study Group ASL parameter recommendations, but most of our cases were scanned before the release of these recommendations. It is possible that the shorter postlabel delay has influenced the findings and observations with the NDP map. These findings can be further validated and confirmed at longer postlabel delay times in future studies.⁵¹