Acute stroke: automatic perfusion lesion outlining using level sets.

PURPOSE: To develop a user-independent algorithm for the delineation of hypoperfused tissue on perfusion-weighted images and evaluate its performance relative to a standard threshold method in simulated data, as well as in acute stroke patients. MATERIALS AND METHODS: The study was approved by the local ethics committee, and patients gave written informed consent prior to their inclusion in the study. The algorithm identifies hypoperfused tissue in mean transit time maps by simultaneously minimizing the mean square error between individual and mean perfusion values inside and outside a smooth boundary. In 14 acute stroke patients, volumetric agreement between automated outlines and manual outlines determined in consensus among four neuroradiologists was assessed with Bland-Altman analysis, while spatial agreement was quantified by using lesion overlap relative to mean lesion volume (Dice coefficient). Performance improvement relative to a standard threshold approach was tested with the Wilcoxon signed rank test. RESULTS: The mean difference in lesion volume between automated outlines and manual outlines was -9.0 mL ± 44.5 (standard deviation). The lowest mean volume difference for the threshold approach was -25.8 mL ± 88.2. A significantly higher Dice coefficient was observed with the algorithm (0.71; interquartile range [IQR], 0.42-0.75) compared with the threshold approach (0.50; IQR, 0.27- 0.57; P , .001). The corresponding agreement among experts was 0.79 (IQR, 0.69-0.83). CONCLUSION: The perfusion lesions outlined by the automated algorithm agreed well with those defined manually in consensus by four experts and were superior to those obtained by using the standard threshold approach. This user-independent algorithm may improve the assessment of perfusion images as part of acute stroke treatment. SUPPLEMENTAL MATERIAL: http://radiology.rsna.org/lookup/suppl/doi:10.1148/radiol.13121622/-/DC1.

Predicting tissue outcome from acute stroke magnetic resonance imaging: improving model performance by optimal sampling of training data.

BACKGROUND AND PURPOSE: It has been hypothesized that algorithms predicting the final outcome in acute ischemic stroke may provide future tools for identifying salvageable tissue and hence guide individualized therapy. We developed means of quantifying predictive model performance to identify model training strategies that optimize performance and reduce bias in predicted lesion volumes. METHODS: We optimized predictive performance based on the area under the receiver operating curve for logistic regression and used simulated data to illustrate the effect of an unbalanced (unequal number of infarcting and surviving voxels) training set on predicted infarct risk. We then tested the performance and optimality of models based on perfusion-weighted, diffusion-weighted, and structural MRI modalities by changing the proportion of mismatch voxels in balanced training material. RESULTS: Predictive performance (area under the receiver operating curve) based on all brain voxels is excessively optimistic and lacks sensitivity in performance in mismatch tissue. The ratio of infarcting and noninfarcting voxels used for training predictive algorithms significantly biases tissue infarct risk estimates. Optimal training strategy is obtained using a balanced training set. We show that 60% of noninfarcted voxels consists of mismatch voxels in an optimal balanced training set for the patient data presented. CONCLUSIONS: An equal number of infarcting and noninfarcting voxels should be used when training predictive models. The choice of test and training sets critically affects predictive model performance and should be closely evaluated before comparisons across patient cohorts.

Predicting tissue outcome in stroke: new approaches.

PURPOSE OF REVIEW: Multimodal MRI provides powerful tools to study acute stroke pathophysiology and to guide stroke therapy. In particular, the perfusion-diffusion mismatch has been hypothesized as a target for treatment beyond the 3 h time window. Studies of infarct progression and of tissue oxygen metabolism suggest that infarct risk is extremely heterogeneous across the diffusion and perfusion lesion. The review describes techniques to more accurately image and model penumbral infarct risk. RECENT FINDINGS: Methods assessing oxygen supply by either blood oxygen level-dependent contrast MRI or models of oxygen delivery capacity may improve the detection of tissue-at-risk. Informatics approaches integrate acute multimodal and follow-up images from large patient cohorts into models of infarct progression. When applied to subsequent acute image data, these techniques may assign infarct risks to mismatch tissue. Recent studies suggest that such estimates of tissue infarct risk may detect treatment-related risk reduction in small patient cohorts. SUMMARY: MRI methods may detect markers of metabolic derangement in ischemia, facilitating the detection of penumbral tissue. Predictive models extend the current perfusion-diffusion mismatch concept by estimating voxel-based risk estimates. With future developments, predictive models may support advanced prognostic support and cost-effective testing of novel stroke therapies.

The role of the cerebral capillaries in acute ischemic stroke: the extended penumbra model.

The pathophysiology of cerebral ischemia is traditionally understood in relation to reductions in cerebral blood flow (CBF). However, a recent reanalysis of the flow-diffusion equation shows that increased capillary transit time heterogeneity (CTTH) can reduce the oxygen extraction efficacy in brain tissue for a given CBF. Changes in capillary morphology are typical of conditions predisposing to stroke and of experimental ischemia. Changes in capillary flow patterns have been observed by direct microscopy in animal models of ischemia and by indirect methods in humans stroke, but their metabolic significance remain unclear. We modeled the effects of progressive increases in CTTH on the way in which brain tissue can secure sufficient oxygen to meet its metabolic needs. Our analysis predicts that as CTTH increases, CBF responses to functional activation and to vasodilators must be suppressed to maintain sufficient tissue oxygenation. Reductions in CBF, increases in CTTH, and combinations thereof can seemingly trigger a critical lack of oxygen in brain tissue, and the restoration of capillary perfusion patterns therefore appears to be crucial for the restoration of the tissue oxygenation after ischemic episodes. In this review, we discuss the possible implications of these findings for the prevention, diagnosis, and treatment of acute stroke.

Cortical hypometabolism and hypoperfusion in Parkinson's disease is extensive: probably even at early disease stages.

Recent cerebral blood flow (CBF) and glucose consumption (CMRglc) studies of Parkinson's disease (PD) revealed conflicting results. Using simulated data, we previously demonstrated that the often-reported subcortical hypermetabolism in PD could be explained as an artifact of biased global mean (GM) normalization, and that low-magnitude, extensive cortical hypometabolism is best detected by alternative data-driven normalization methods. Thus, we hypothesized that PD is characterized by extensive cortical hypometabolism but no concurrent widespread subcortical hypermetabolism and tested it on three independent samples of PD patients. We compared SPECT CBF images of 32 early-stage and 33 late-stage PD patients with that of 60 matched controls. We also compared PET FDG images from 23 late-stage PD patients with that of 13 controls. Three different normalization methods were compared: (1) GM normalization, (2) cerebellum normalization, (3) reference cluster normalization (Yakushev et al.). We employed standard voxel-based statistics (fMRIstat) and principal component analysis (SSM). Additionally, we performed a meta-analysis of all quantitative CBF and CMRglc studies in the literature to investigate whether the global mean (GM) values in PD are decreased. Voxel-based analysis with GM normalization and the SSM method performed similarly, i.e., both detected decreases in small cortical clusters and concomitant increases in extensive subcortical regions. Cerebellum normalization revealed more widespread cortical decreases but no subcortical increase. In all comparisons, the Yakushev method detected nearly identical patterns of very extensive cortical hypometabolism. Lastly, the meta-analyses demonstrated that global CBF and CMRglc values are decreased in PD. Based on the results, we conclude that PD most likely has widespread cortical hypometabolism, even at early disease stages. In contrast, extensive subcortical hypermetabolism is probably not a feature of PD.

Normalization in PET group comparison studies--the importance of a valid reference region.

INTRODUCTION: In positron emission tomography (PET) studies of cerebral blood flow (CBF) and metabolism, the large interindividual variation commonly is minimized by normalization to the global mean prior to statistical analysis. This approach requires that no between-group or between-state differences exist in the normalization region. Given the variability typical of global CBF and the practical limit on sample size, small group differences in global mean easily elude detection, but still bias the comparison, with profound consequences for the physiological interpretation of the results. MATERIALS AND METHODS: Quantitative [15O]H2O PET recordings of CBF were obtained in 45 healthy subjects (21-81 years) and 14 patients with hepatic encephalopathy (HE). With volume-of-interest (VOI) and voxel-based statistics, we conducted regression analyses of CBF as function of age in the healthy group, and compared the HE group to a subset of the controls. We compared absolute CBF values, and CBF normalized to the gray matter (GM) and white matter (WM) means. In additional simulation experiments, we manipulated the cortical values of 12 healthy subjects and compared these to unaltered control data. RESULTS: In healthy aging, CBF was shown to be unchanged in WM and central regions. In contrast, with normalization to the GM mean, CBF displayed positive correlation with age in the central regions. Very similar artifactual increases were seen in the HE comparison and also in the simulation experiment. CONCLUSION: Ratio normalization to the global mean readily elevates CBF in unchanged regions when a systematic between-group difference exists in gCBF, also when this difference is below the detection threshold. We suggest that the routine normalization to the global mean in earlier studies resulted in spurious interpretations of perturbed CBF. Normalization to central WM yields less biased results in aging and HE and could potentially serve as a normalization reference region in other disorders as well.