Clinical application of computed-tomography based imaging in acute ischemic stroke

Ischemic stroke is a devastating disease and one of the leading causes of disability. Blood vessel occlusion in ischemic stroke limits blood flow to downstream brain tissue. Severe hypoperfusion causes complete cellular energy depletion, resulting in cell necrosis in an area called the ischemic core. Less severely hypoperfused brain tissue surrounding the ischemic core contributes to the patient’s symptoms, but remains viable upon rapid restoration of perfusion and is termed the penumbra. If perfusion is not restored, cellular compensation mechanisms in the penumbra will ultimately fail and irreversible damage will ensue, thus expanding the ischemic core. The rate of ischemic core growth is variable between individuals, largely explained by differences in collateral circulation, i.e. a network of blood vessel connections that maintain sufficient perfusion to ischemic brain tissue in case of upstream blood vessel occlusion and thereby temporarily prevent irreversible damage.

Acute therapy for ischemic stroke aims to rapidly restore blood flow to penumbral tissue, ideally preventing any irreversible brain damage. The effect of reperfusion treatment using either thrombolytic therapy (thrombolysis) or endovascular therapy (EVT) decreases over time due to failure of collateral circulation. Individual treatment effect may however persist well beyond conventional treatment time windows validated on the population level because of variability in collateral circulation quality.

Brain imaging is a cornerstone of acute stroke treatment to exclude hemorrhagic lesions and large areas of subacute ischemic injury. Non-contrast computed tomography (NCCT) is very sensitive for acute hemorrhage, readily available, fast and cheap and therefore widely used as the first imaging modality in the diagnostic evaluation of presumed stroke. Signal-to-noise ratio and sensitivity for early signs of ischemia are however low, resulting in considerable interrater variability. In an attempt to quantify early ischemic changes on NCCT, the Alberta Stroke Program Early CT Score (ASPECTS) was developed. ASPECTS scores early ischemic changes in 7 cortical and 3 deep brain regions and has shown to predict functional outcome after reperfusion treatment for acute ischemic stroke.

Contrary to structural imaging like NCCT, perfusion imaging is able to not only estimate established infarction, but also predict tissue at risk of infarction by quantifying contrast passage through the brain parenchyma over time. CT perfusion (CTP) is increasingly used in acute ischemic stroke. In CTP, iodine contrast is injected and the resulting attenuation of brain tissue voxels is measured over time. From the resulting time x tissue-concentration curve, several perfusion parameters can be deducted. Although these parameters do not measure infarction directly, they are able to predict brain tissue fate. Cerebral blood flow, deducted from the slope of the tissue-concentration curve, less than 30% relative to non-ischemic brain tissue (relative CBF, rCBF) accurately predicts the ischemic core. Time from the first deflection to the maximum deflection of the tissue-concentration curve (time-to-maximum, Tmax) more than 6 seconds correlates well with the brain volume at risk of infarction in the absence of reperfusion (i.e., the perfusion lesion). Several observational studies have shown that a mismatch between the predicted ischemic core and the perfusion lesion, i.e., a limited core and the presence of a considerable penumbra, is associated with response to reperfusion therapy whereas absence of this mismatch may not be. This finding was later used to select patients for randomized, controlled trials on thrombolysis and EVT vs. best medical treatment in late or unknown treatment time windows. These trials showed demonstrated benefit of reperfusion in mismatch-selected patients.

In clinical practice, both structural imaging alone and additional perfusion imaging are variably used for diagnosis and treatment selection of acute ischemic stroke, but comparison of these imaging techniques was largely lacking. In this thesis, we aimed to compare the use of NCCT and CTP and explore measures to overcome limitations of both imaging modalities.

We demonstrated that both the CTP core and ASPECTS accurately discriminate smaller (< 70 mL) from larger (≥ 70 mL) acute infarcts on near-simultaneous MRI in a population of acute anterior circulation ischemic stroke patients. The accuracy of ASPECTS was however numerically lower which translates to a 12% excess of incorrectly classified acute stroke patients compared to using CTP core measurement. Nonetheless, we found a systematic underestimation of the infarct volume using CTP which was greater in patients who did not achieve reperfusion at the time of MRI, likely representing ongoing infarct growth. Acute infarct volume is associated with functional outcome and may thus be a relevant prognostic marker. Moreover, recent evidence from the RESCUE-Japan Limit trial, which randomized patients with acute anterior circulation ischemic stroke and low ASPECTS to EVT vs. best medical therapy, demonstrates that patients with very low ASPECTS (0-3) may no longer benefit from EVT. Nonetheless, infarct volumes in the group that did experience EVT benefit were greater (median 89 mL) than the cut-off used in our study. Future studies in patients with large infarcts will hopefully further elucidate the boundaries  of therapeutic effectiveness.

We also demonstrated that CTP core volume is associated with functional outcome in a population of acute ischemic stroke with complete endovascular reperfusion of a large vessel occlusion within 18 hours from symptom onset or time last known well. This association remained after correction for relevant clinical confounders. Predictive accuracy was however low, likely resulting from limited variability in ischemic core volumes in the study population (core volume > 50 mL in only 10% of included patients). Contrary to the association between CTP core and functional outcome, we did not find an association between ASPECTS and functional outcome. To overcome interrater variability with ASPECTS, we performed the same analysis with a software-generated, automated ASPECTS (Brainomix eASPECTS). Similarly to conventional ASPECTS, automated ASPECTS was associated with good functional outcome only after adjustment for confounders, but not with poor functional outcome. Although variability in both ischemic core volumes and ASPECTS was low, these results seem to underpin the superiority of CTP core for predication of outcome in reperfused acute ischemic stroke, and simultaneously highlight that other clinical and imaging variables strongly contribute to functional outcome, while the predictive capacity of ischemic core volume or ASPECTS alone is limited. In the prospective observational FRAME study, which included patients with anterior circulation large vessel occlusion who underwent EVT irrespective of perfusion imaging results, EVT effect was demonstrated in patients with a penumbra of at least 10 mL irrespective of ischemic core volume, but not in patients without penumbra.

Other imaging parameters may also be used to predict functional outcome. In a population of patients who underwent EVT within 18 hours from last known well for acute ischemic stroke due to anterior circulation large vessel occlusion, we demonstrated that women had better collateral circulation and likely therefore less ischemic core growth, smaller final infarct volumes and better functional outcomes. Collateral circulation was measured indirectly by CTP using the hypoperfusion intensity ratio, which quantifies the proportion of the perfusion lesion that is severely hypoperfused (Tmax > 10 seconds / Tmax >6 seconds). Although other studies have confirmed better collaterals in women, the association with sex may have been coincidental, since outcome differences between men and women have not consistently been shown.

Our results uncovered problematic ASPECTS interrater variability. Although weighted measures of agreement showed a fair to good agreement, absolute agreement was low, causing problems when ASPECTS is used to identify candidates for acute reperfusion therapy. We also demonstrated that correlation between CTP core volume and both conventional ASPECTS and automated ASPECTS was weak. These findings are in line with other studies that show a large variability of ischemic core volume per ASPECTS strata. Our findings have been validated in large observational studies where performance of raters of variable experience was compared. The results of these studies suggest that ASPECTS interrater agreement may be insufficient to justify using ASPECTS for patient selection, even by experienced raters.  

To overcome ASPECTS interrater variability and insufficient correlation with infarct size, we propose a semi-automated NCCT lesion segmentation tool based on hypoattenuation relative to the corresponding contralateral brain region. Derived in a population of late-presenting acute ischemic stroke patients with anterior circulation large vessel occlusion, an optimal threshold of 5.1% Hounsfield units depression detected the least amount of false positive lesion volume while preserving detection of true positive infarct volume when compared to co-registered follow-up imaging. We were able to show that rNCCT-assisted segmentation improved interrater agreement between experienced as well as inexperienced raters, while simultaneously increasing the detection yield of ischemic regions compared to unassisted NCCT segmentations. We believe this tool could aid rapid, and likely more accurate, detection of ischemic changes on NCCT in clinical practice. However, the value of rNCCT for prediction of outcome and patient selection remains undetermined.

Similar to ASPECTS and other unassisted NCCT evaluation methods, CTP also has several limitations. Other studies have e.g. shown that clinically used ischemic core thresholds (i.e., rCBF < 30%) overestimate the area of infarction in a non-negligible proportion of patients who present very early after stroke onset. In those patients, a more restrictive rCBF < 20% threshold has been proposed. In a mixed population early and late presenting acute ischemic stroke patients with complete endovascular reperfusion, we confirmed this finding and demonstrated that CTP may also underestimate the ischemic core, especially in patients who present late after symptom onset or time last known well. Compared to co-registered follow-up diffusion weighted MRI, we found a median CTP core underestimation of 11 mL, and 17 mL in late presenters, which ranged up to 148 mL in individual cases. This finding is likely not entirely explained by continued infarct growth, since only patients with complete endovascular recanalization were selected and since the proportion of underestimation is greater in late vs. early presenters whereas the time interval between imaging and reperfusion was not. By performing  rNCCT segmentations coregistered to baseline CTP and follow-up diffusion-weighted MRI, we showed that median 24% of the infarction undercalled by CTP core could be detected by NCCT. Furthermore, we showed that combining the rNCCT and CTP core lesion increased the accuracy for detection of the follow-up infarct. We thereby provided irrefutable evidence that NCCT and CTP contain complementary information on tissue status.

In conclusion, both NCCT and CTP core volume provide important information regarding brain tissue status in patients with acute ischemic stroke. Although an ASPECTS cut-off can be used to estimate the acute infarct volume, there are many drawbacks to this imaging method. We believe that (semi-) automated NCCT infarct segmentation is a promising tool that can be used complementary to CTP in an effort to further improve prognostication in acute ischemic stroke, and perhaps optimize identification of patients with reperfusion treatment targets.