OBJECTIVE: Most cardiac arrest patients who are successfully resuscitated are initially comatose due to hypoxic-ischemic brain injury. Quantitative electroencephalography (EEG) provides valuable prognostic information. However, prior approaches largely rely on snapshots of the EEG, without taking advantage of temporal information. METHODS: We present a recurrent deep neural network with the goal of capturing temporal dynamics from longitudinal EEG data to predict long-term neurological outcomes. We utilized a large international dataset of continuous EEG recordings from 1,038 cardiac arrest patients from seven hospitals in Europe and the US. Poor outcome was defined as a Cerebral Performance Category (CPC) score of 3-5, and good outcome as CPC score 0-2 at 3 to 6-months after cardiac arrest. Model performance is evaluated using 5-fold cross validation. RESULTS: The proposed approach provides predictions which improve over time, beginning from an area under the receiver operating characteristic curve (AUC-ROC) of 0.78 (95% CI: 0.72-0.81) at 12 hours, and reaching 0.88 (95% CI: 0.85-0.91) by 66 h after cardiac arrest. At 66 h, (sensitivity, specificity) points of interest on the ROC curve for predicting poor outcomes were (32,99)%, (55,95)%, and (62,90)%, (99,23)%, (95,47)%, and (90,62)%; whereas for predicting good outcome, the corresponding operating points were (17,99)%, (47,95)%, (62,90)%, (99,19)%, (95,48)%, (70,90)%. Moreover, the model provides predicted probabilities that closely match the observed frequencies of good and poor outcomes (calibration error 0.04). CONCLUSIONS AND SIGNIFICANCE: These findings suggest that accounting for EEG trend information can substantially improve prediction of neurologic outcomes for patients with coma following cardiac arrest.

OBJECTIVE: Electroencephalography (EEG) is an important tool for neurological outcome prediction after cardiac arrest. However, the complexity of continuous EEG data limits timely and accurate interpretation by clinicians. We develop a deep neural network (DNN) model to leverage complex EEG trends for early and accurate assessment of cardiac arrest coma recovery likelihood. METHODS: We developed a multiscale DNN combining convolutional neural networks (CNN) and recurrent neural networks (long short-term memory [LSTM]) using EEG and demographic information (age, gender, shockable rhythm) from a multicenter cohort of 1,038 cardiac arrest patients. The CNN learns EEG feature representations while the multiscale LSTM captures short-term and long-term EEG dynamics on multiple time scales. Poor outcome is defined as a Cerebral Performance Category (CPC) score of 3-5 and good outcome as CPC score 1-2 at 3-6 months after cardiac arrest. Performance is evaluated using area under the receiver operating characteristic curve (AUC) and calibration error. RESULTS: Model performance increased with EEG duration, with AUC increasing from 0.83 (95% Confidence Interval [CI] 0.79-0.87 at 12h to 0.91 (95%CI 0.88-0.93) at 66h. Sensitivity of good and poor outcome prediction was 77% and 75% at a specificity of 90%, respectively. Sensitivity of poor outcome was 50% at a specificity of 99%. Predicted probability was well matched to the observation frequency of poor outcomes, with a calibration error of 0.11 [0.09-0.14]. CONCLUSIONS: These results demonstrate that incorporating EEG evolution over time improves the accuracy of neurologic outcome prediction for patients with coma after cardiac arrest.

BACKGROUND AND PURPOSE: Accurate automated infarct segmentation is needed for acute ischemic stroke studies relying on infarct volumes as an imaging phenotype or biomarker that require large numbers of subjects. This study investigated whether an ensemble of convolutional neural networks trained on multiparametric DWI maps outperforms single networks trained on solo DWI parametric maps. MATERIALS AND METHODS: Convolutional neural networks were trained on combinations of DWI, ADC, and low b-value-weighted images from 116 subjects. The performances of the networks (measured by the Dice score, sensitivity, and precision) were compared with one another and with ensembles of 5 networks. To assess the generalizability of the approach, we applied the best-performing model to an independent Evaluation Cohort of 151 subjects. Agreement between manual and automated segmentations for identifying patients with large lesion volumes was calculated across multiple thresholds (21, 31, 51, and 70 cm). RESULTS: An ensemble of convolutional neural networks trained on DWI, ADC, and low b-value-weighted images produced the most accurate acute infarct segmentation over individual networks ( < .001). Automated volumes correlated with manually measured volumes (Spearman ρ = 0.91, < .001) for the independent cohort. For the task of identifying patients with large lesion volumes, agreement between manual outlines and automated outlines was high (Cohen κ, 0.86-0.90; < .001). CONCLUSIONS: Acute infarcts are more accurately segmented using ensembles of convolutional neural networks trained with multiparametric maps than by using a single model trained with a solo map. Automated lesion segmentation has high agreement with manual techniques for identifying patients with large lesion volumes.

Background and Purpose- We evaluated deep learning algorithms' segmentation of acute ischemic lesions on heterogeneous multi-center clinical diffusion-weighted magnetic resonance imaging (MRI) data sets and explored the potential role of this tool for phenotyping acute ischemic stroke. Methods- Ischemic stroke data sets from the MRI-GENIE (MRI-Genetics Interface Exploration) repository consisting of 12 international genetic research centers were retrospectively analyzed using an automated deep learning segmentation algorithm consisting of an ensemble of 3-dimensional convolutional neural networks. Three ensembles were trained using data from the following: (1) 267 patients from an independent single-center cohort, (2) 267 patients from MRI-GENIE, and (3) mixture of (1) and (2). The algorithms' performances were compared against manual outlines from a separate 383 patient subset from MRI-GENIE. Univariable and multivariable logistic regression with respect to demographics, stroke subtypes, and vascular risk factors were performed to identify phenotypes associated with large acute diffusion-weighted MRI volumes and greater stroke severity in 2770 MRI-GENIE patients. Stroke topography was investigated. Results- The ensemble consisting of a mixture of MRI-GENIE and single-center convolutional neural networks performed best. Subset analysis comparing automated and manual lesion volumes in 383 patients found excellent correlation (ρ=0.92; P<0.0001). Median (interquartile range) diffusion-weighted MRI lesion volumes from 2770 patients were 3.7 cm (0.9-16.6 cm). Patients with small artery occlusion stroke subtype had smaller lesion volumes ( P<0.0001) and different topography compared with other stroke subtypes. Conclusions- Automated accurate clinical diffusion-weighted MRI lesion segmentation using deep learning algorithms trained with multi-center and diverse data is feasible. Both lesion volume and topography can provide insight into stroke subtypes with sufficient sample size from big heterogeneous multi-center clinical imaging phenotype data sets.

BACKGROUND AND PURPOSE: Tissue signatures from acute MR imaging of the brain may be able to categorize physiological status and thereby assist clinical decision making. We designed and analyzed statistical algorithms to evaluate the risk of infarction for each voxel of tissue using acute human functional MRI. METHODS: Diffusion-weighted MR images (DWI) and perfusion-weighted MR images (PWI) from acute stroke patients scanned within 12 hours of symptom onset were retrospectively studied and used to develop thresholding and generalized linear model (GLM) algorithms predicting tissue outcome as determined by follow-up MRI. The performances of the algorithms were evaluated for each patient by using receiver operating characteristic curves. RESULTS: At their optimal operating points, thresholding algorithms combining DWI and PWI provided 66% sensitivity and 83% specificity, and GLM algorithms combining DWI and PWI predicted with 66% sensitivity and 84% specificity voxels that proceeded to infarct. Thresholding algorithms that combined DWI and PWI provided significant improvement to algorithms that utilized DWI alone (P=0.02) but no significant improvement over algorithms utilizing PWI alone (P=0.21). GLM algorithms that combined DWI and PWI showed significant improvement over algorithms that used only DWI (P=0.02) or PWI (P=0.04). The performances of thresholding and GLM algorithms were comparable (P>0.2). CONCLUSIONS: Algorithms that combine acute DWI and PWI can assess the risk of infarction with higher specificity and sensitivity than algorithms that use DWI or PWI individually. Methods for quantitatively assessing the risk of infarction on a voxel-by-voxel basis show promise as techniques for investigating the natural spatial evolution of ischemic damage in humans.

OBJECTIVE: Acute cerebral stroke is a frequent cause of death and the major cause of adult neurological disability in the western world. Thrombolysis is the only established treatment of ischemic stroke; however, its use carries a substantial risk of symptomatic intracerebral hemorrhage. A clinical tool to guide the use of thrombolysis would be very valuable. One of the major goals of such a tool would be the identification of potentially salvageable tissue. This requires an accurate prediction of the extent of infarction if untreated. In this study, we investigate the applicability of highly flexible instance-based (IB) methods for such predictions. METHODS AND MATERIALS: Based on information obtained from magnetic resonance imaging of 14 patients with acute stroke, we explored three different implementations of the IB method: k-NN, Gaussian weighted, and constant radius search classification. Receiver operating characteristics analysis, in particular area under the curve (AUC), was used as performance measure. RESULTS: We found no significant difference (P = 0.48) in performance for the optimal k-NN (k = 164, AUC = 0.814 +/- 0.001) and Gaussian weight (sigma = 0.17, AUC = 0.813 +/- 0.001) implementations, while they were both significantly better (P < 1 x 10(-6) for both) than the constant radius implementation (R = 0.28, AUC = 0.809 +/- 0.001). Qualitative analyses of the distribution of instances in the feature space indicated that non-infarcted instances tends to cluster together while infarcted instances are more dispersed, and that there may not exist a stringent boundary separating infarcted from non-infarcted instances. CONCLUSIONS: This study shows that IB methods can be used, and may be advantageous, for predicting final infarct in patients with acute stroke, but further work must be done to make them clinically applicable.

Magnetic resonance imaging (MRI) has been shown to improve the diagnosis and management of patients with brain disorders. Multiparametric MRI offers the possibility of noninvasively assessing multiple facets of pathophysiological processes that exist simultaneously, thereby further assisting in patient treatment management. Voxel-based analysis approaches, such as tissue theme mapping, have the benefit over volumetric approaches in being able to identify spatially heterogeneous colocalized changes on multiple parametric MR images that are not readily discernible. Tissue theme maps seem to be a promising tool for integrating the plethora of novel imaging contrasts that are being developed for the noninvasive investigation of the different stages of disease progression into easily interpretable maps of brain injury. We describe here various implementations for combining multiparametric imaging and their merits in the evaluation of brain diseases.

PURPOSE: To examine whether the severity and spatial distribution of reductions in apparent diffusion coefficient (ADC) are associated with clinical outcomes in patients who become comatose after cardiac arrest. MATERIALS AND METHODS: This was an institutional review board-approved, HIPAA-compliant retrospective study of 80 comatose patients with cardiac arrest who underwent diffusion-weighted magnetic resonance imaging. The need to obtain informed consent was waived except when follow-up phone calls were required; in those cases, informed consent was obtained from the families. Mean patient age was 57 years +/- 16 (standard deviation); 31 (39%) patients were women. ADC maps were semiautomatically segmented into the following regions: subcortical white matter; cerebellum; insula; frontal, occipital, parietal, and temporal lobes; caudate nucleus; putamen; and thalamus. Median ADCs were measured in these regions and in the whole brain and were compared (with a two-tailed Wilcoxon test) as a function of clinical outcome. Outcome was defined by both early eye opening in the 1st week after arrest (either spontaneously or in response to external stimuli) and 6-month modified Rankin scale score. RESULTS: Whole-brain median ADC was a significant predictor of poor outcome as measured by no eye opening (specificity, 100% [95% confidence interval {CI}: 86%, 100%]; sensitivity, 30% [95% CI: 18%, 45%]) or 6-month modified Rankin scale score greater than 3 (specificity, 100% [95% CI: 73%, 100%]; sensitivity, 41% [95% CI: 29%, 54%]), with patients with poor outcomes having significantly lower ADCs for both outcome measures (P

We propose and demonstrate an inference algorithm for the automatic segmentation of cerebrovascular pathologies in clinical MR images of the brain. Identifying and differentiating pathologies is important for understanding the underlying mechanisms and clinical outcomes of cerebral ischemia. Manual delineation of separate pathologies is infeasible in large studies of stroke that include thousands of patients. Unlike normal brain tissues and structures, the location and shape of the lesions vary across patients, presenting serious challenges for prior-driven segmentation. Our generative model captures spatial patterns and intensity properties associated with different cerebrovascular pathologies in stroke patients. We demonstrate the resulting segmentation algorithm on clinical images of a stroke patient cohort.

There is increasing interest in using algorithms combining multiple magnetic resonance imaging (MRI) modalities to predict tissue infarction in acute human stroke. We developed and tested a voxel-based generalized linear model (GLM) algorithm to predict tissue infarction in an animal stroke model in order to directly compare predicted outcome with the tissue's histologic outcome, and to evaluate the potential for assessing therapeutic efficacy using these multiparametric algorithms. With acute MRI acquired after unilateral embolic stroke in rats (n=8), a GLM was developed and used to predict infarction on a voxel-wise basis for saline (n=6) and recombinant tissue plasminogen activator (rt-PA) treatment (n=7) arms of a trial of delayed thrombolytic therapy in rats. Pretreatment predicted outcome compared with post-treatment histology was highly accurate in saline-treated rats (0.92+/-0.05). Accuracy was significantly reduced (P=0.04) in rt-PA-treated animals (0.86+/-0.08), although no significant difference was detected when comparing histologic lesion volumes. Animals that reperfused had significantly lower (P<0.01) GLM-predicted infarction risk (0.73+/-0.03) than nonreperfused animals (0.81+/-0.05), possibly reflecting less severe initial ischemic injury and therefore tissue likely more amenable to therapy. Our results show that acute MRI-based algorithms can predict tissue infarction with high accuracy in animals not receiving thrombolytic therapy. Furthermore, alterations in disease progression due to treatment were more sensitively monitored with our voxel-based analysis techniques than with volumetric approaches. Our study shows that predictive algorithms are promising metrics for diagnosis, prognosis and therapeutic evaluation after acute stroke that can translate readily from preclinical to clinical settings.