Publications

    Winzeck S, Mocking SJT, Bezerra R, Bouts MJRJ, McIntosh EC, Diwan I, Garg P, Chutinet A, Kimberly WT, Copen WA, Schaefer PW, Ay H, Singhal AB, Kamnitsas K, Glocker B, Sorensen AG, Wu O. Ensemble of Convolutional Neural Networks Improves Automated Segmentation of Acute Ischemic Lesions Using Multiparametric Diffusion-Weighted MRI. AJNR Am J Neuroradiol 2019;40(6):938-945.Abstract
    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.
    Wu O, Winzeck S, Giese A-K, Hancock BL, Etherton MR, Bouts MJRJ, Donahue K, Schirmer MD, Irie RE, Mocking SJT, McIntosh EC, Bezerra R, Kamnitsas K, Frid P, Wasselius J, Cole JW, Xu H, Holmegaard L, Jiménez-Conde J, Lemmens R, Lorentzen E, McArdle PF, Meschia JF, Roquer J, Rundek T, Sacco RL, Schmidt R, Sharma P, Slowik A, Stanne TM, Thijs V, Vagal A, Woo D, Bevan S, Kittner SJ, Mitchell BD, Rosand J, Worrall BB, Jern C, Lindgren AG, Maguire J, Rost NS. Big Data Approaches to Phenotyping Acute Ischemic Stroke Using Automated Lesion Segmentation of Multi-Center Magnetic Resonance Imaging Data. Stroke 2019;50(7):1734-1741.Abstract
    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.
    Bouts MJRJ, Tiebosch IA, Rudrapatna US, van der Toorn A, Wu O, Dijkhuizen RM. Prediction of hemorrhagic transformation after experimental ischemic stroke using MRI-based algorithms. J Cereb Blood Flow Metab 2017;37(8):3065-3076.Abstract
    Estimation of hemorrhagic transformation (HT) risk is crucial for treatment decision-making after acute ischemic stroke. We aimed to determine the accuracy of multiparametric MRI-based predictive algorithms in calculating probability of HT after stroke. Spontaneously, hypertensive rats were subjected to embolic stroke and, after 3 h treated with tissue plasminogen activator (Group I: n = 6) or vehicle (Group II: n = 7). Brain MRI measurements of T2, T2*, diffusion, perfusion, and blood-brain barrier permeability were obtained at 2, 24, and 168 h post-stroke. Generalized linear model and random forest (RF) predictive algorithms were developed to calculate the probability of HT and infarction from acute MRI data. Validation against seven-day outcome on MRI and histology revealed that highest accuracy of hemorrhage prediction was achieved with a RF-based model that included spatial brain features (Group I: area under the receiver-operating characteristic curve (AUC) = 0.85 ± 0.14; Group II: AUC = 0.89 ± 0.09), with significant improvement over perfusion- or permeability-based thresholding methods. However, overlap between predicted and actual tissue outcome was significantly lower for hemorrhage prediction models (maximum Dice's Similarity Index (DSI) = 0.20 ± 0.06) than for infarct prediction models (maximum DSI = 0.81 ± 0.06). Multiparametric MRI-based predictive algorithms enable early identification of post-ischemic tissue at risk of HT and may contribute to improved treatment decision-making after acute ischemic stroke.
    Bouts MJRJ, Westmoreland SV, de Crespigny AJ, Liu Y, Vangel M, Dijkhuizen RM, Wu O, D'Arceuil HE. Magnetic resonance imaging-based cerebral tissue classification reveals distinct spatiotemporal patterns of changes after stroke in non-human primates. BMC Neurosci 2015;16:91.Abstract
    BACKGROUND: Spatial and temporal changes in brain tissue after acute ischemic stroke are still poorly understood. Aims of this study were three-fold: (1) to determine unique temporal magnetic resonance imaging (MRI) patterns at the acute, subacute and chronic stages after stroke in macaques by combining quantitative T2 and diffusion MRI indices into MRI 'tissue signatures', (2) to evaluate temporal differences in these signatures between transient (n = 2) and permanent (n = 2) middle cerebral artery occlusion, and (3) to correlate histopathology findings in the chronic stroke period to the acute and subacute MRI derived tissue signatures. RESULTS: An improved iterative self-organizing data analysis algorithm was used to combine T2, apparent diffusion coefficient (ADC), and fractional anisotropy (FA) maps across seven successive timepoints (1, 2, 3, 24, 72, 144, 240 h) which revealed five temporal MRI signatures, that were different from the normal tissue pattern (P < 0.001). The distribution of signatures between brains with permanent and transient occlusions varied significantly between groups (P < 0.001). Qualitative comparisons with histopathology revealed that these signatures represented regions with different histopathology. Two signatures identified areas of progressive injury marked by severe necrosis and the presence of gitter cells. Another signature identified less severe but pronounced neuronal and axonal degeneration, while the other signatures depicted tissue remodeling with vascular proliferation and astrogliosis. CONCLUSION: These exploratory results demonstrate the potential of temporally and spatially combined voxel-based methods to generate tissue signatures that may correlate with distinct histopathological features. The identification of distinct ischemic MRI signatures associated with specific tissue fates may further aid in assessing and monitoring the efficacy of novel pharmaceutical treatments for stroke in a pre-clinical and clinical setting.
    Schaefer PW, Pulli B, Copen WA, Hirsch JA, Leslie-Mazwi T, Schwamm LH, Wu O, González RG, Yoo AJ. Combining MRI with NIHSS thresholds to predict outcome in acute ischemic stroke: value for patient selection. AJNR Am J Neuroradiol 2015;36(2):259-64.Abstract
    BACKGROUND AND PURPOSE: Selecting acute ischemic stroke patients for reperfusion therapy on the basis of a diffusion-perfusion mismatch has not been uniformly proved to predict a beneficial treatment response. In a prior study, we have shown that combining clinical with MR imaging thresholds can predict clinical outcome with high positive predictive value. In this study, we sought to validate this predictive model in a larger patient cohort and evaluate the effects of reperfusion therapy and stroke side. MATERIALS AND METHODS: One hundred twenty-three consecutive patients with anterior circulation acute ischemic stroke underwent MR imaging within 6 hours of stroke onset. DWI and PWI volumes were measured. Lesion volume and NIHSS score thresholds were used in models predicting good 3-month clinical outcome (mRS 0-2). Patients were stratified by treatment and stroke side. RESULTS: Receiver operating characteristic analysis demonstrated 95.6% and 100% specificity for DWI > 70 mL and NIHSS score > 20 to predict poor outcome, and 92.7% and 91.3% specificity for PWI (mean transit time) < 50 mL and NIHSS score < 8 to predict good outcome. Combining clinical and imaging thresholds led to an 88.8% (71/80) positive predictive value with a 65.0% (80/123) prognostic yield. One hundred percent specific thresholds for DWI (103 versus 31 mL) and NIHSS score (20 versus 17) to predict poor outcome were significantly higher in treated (intravenous and/or intra-arterial) versus untreated patients. Prognostic yield was lower in right- versus left-sided strokes for all thresholds (10.4%-20.7% versus 16.9%-40.0%). Patients with right-sided strokes had higher 100% specific DWI (103.1 versus 74.8 mL) thresholds for poor outcome, and the positive predictive value was lower. CONCLUSIONS: Our predictive model is validated in a much larger patient cohort. Outcome may be predicted in up to two-thirds of patients, and thresholds are affected by stroke side and reperfusion therapy.
    Dalca AV, Sridharan R, Cloonan L, Fitzpatrick KM, Kanakis A, Furie KL, Rosand J, Wu O, Sabuncu M, Rost NS, Golland P. Segmentation of cerebrovascular pathologies in stroke patients with spatial and shape priors. Med Image Comput Comput Assist Interv 2014;17(Pt 2):773-80.Abstract
    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.
    Bouts MJRJ, Tiebosch IA, van der Toorn A, Viergever MA, Wu O, Dijkhuizen RM. Early identification of potentially salvageable tissue with MRI-based predictive algorithms after experimental ischemic stroke. J Cereb Blood Flow Metab 2013;33(7):1075-82.Abstract
    Individualized stroke treatment decisions can be improved by accurate identification of the extent of salvageable tissue. Magnetic resonance imaging (MRI)-based approaches, including measurement of a 'perfusion-diffusion mismatch' and calculation of infarction probability, allow assessment of tissue-at-risk; however, the ability to explicitly depict potentially salvageable tissue remains uncertain. In this study, five predictive algorithms (generalized linear model (GLM), generalized additive model, support vector machine, adaptive boosting, and random forest) were tested in their potency to depict acute cerebral ischemic tissue that can recover after reperfusion. Acute T2-, diffusion-, and perfusion-weighted MRI, and follow-up T2 maps were collected from rats subjected to right-sided middle cerebral artery occlusion without subsequent reperfusion, for training of algorithms (Group I), and with spontaneous (Group II) or thrombolysis-induced reperfusion (Group III), to determine infarction probability-based viability thresholds and prediction accuracies. The infarction probability difference between irreversible-i.e., infarcted after reperfusion-and salvageable tissue injury-i.e., noninfarcted after reperfusion-was largest for GLM (20±7%) with highest accuracy of risk-based identification of acutely ischemic tissue that could recover on subsequent reperfusion (Dice's similarity index=0.79±0.14). Our study shows that assessment of the heterogeneity of infarction probability with MRI-based algorithms enables estimation of the extent of potentially salvageable tissue after acute ischemic stroke.
    Wu O, Benner T, Roccatagliata L, Zhu M, Schaefer PW, Sorensen AG, Singhal AB. Evaluating effects of normobaric oxygen therapy in acute stroke with MRI-based predictive models. Med Gas Res 2012;2(1):5.Abstract
    BACKGROUND: Voxel-based algorithms using acute multiparametric-MRI data have been shown to accurately predict tissue outcome after stroke. We explored the potential of MRI-based predictive algorithms to objectively assess the effects of normobaric oxygen therapy (NBO), an investigational stroke treatment, using data from a pilot study of NBO in acute stroke. METHODS: The pilot study of NBO enrolled 11 patients randomized to NBO administered for 8 hours, and 8 Control patients who received room-air. Serial MRIs were obtained at admission, during gas therapy, post-therapy, and pre-discharge. Diffusion/perfusion MRI data acquired at admission (pre-therapy) was used in generalized linear models to predict the risk of lesion growth at subsequent time points for both treatment scenarios: NBO or Control. RESULTS: Lesion volume sizes 'during NBO therapy' predicted by Control-models were significantly larger (P = 0.007) than those predicted by NBO models, suggesting that ischemic lesion growth is attenuated during NBO treatment. No significant difference was found between the predicted lesion volumes at later time-points. NBO-treated patients, despite showing larger lesion volumes on Control-models than NBO-models, tended to have reduced lesion growth. CONCLUSIONS: This study shows that NBO has therapeutic potential in acute ischemic stroke, and demonstrates the feasibility of using MRI-based algorithms to evaluate novel treatments in early-phase clinical trials.
    Wu O, Christensen S, Hjort N, Dijkhuizen RM, Kucinski T, Fiehler J, Thomalla G, Röther J, Østergaard L. Characterizing physiological heterogeneity of infarction risk in acute human ischaemic stroke using MRI. Brain 2006;129(Pt 9):2384-93.Abstract
    Viable tissues at risk of infarction in acute stroke patients have been hypothesized to be detectable as volumetric mismatches between lesions on perfusion-weighted (PWI) and diffusion-weighted magnetic resonance imaging (DWI). Because tissue response to ischaemic injury and to therapeutic intervention is tissue- and patient-dependent, changes in infarct progression due to treatment may be better detected with voxel-based methods than with volumetric mismatches. Acute DWI and PWI were combined using a generalized linear model (GLM) to predict infarction risk on a voxel-wise basis for patients treated either with non-thrombolytic (Group 1; n = 11) or with thrombolytic therapy (Group 2; n = 27). Predicted infarction risk for both groups was evaluated in four ipsilateral regions of interest: tissue acutely abnormal on DWI (Core), tissue acutely abnormal on PWI but normal on DWI that either infarcts (Recruited) or does not (Salvaged), and tissue normal on both DWI and PWI that does not infarct (Normal) by follow-up imaging > or = 5 days. The performance of the models was significantly reduced for the thrombolysed group compared with the group receiving standard treatment, suggesting an alteration in natural progression of the ischaemic cascade. Average GLM-predicted infarction risk values in the four regions were different from one another for both groups. GLM-predicted infarction risk in Salvaged tissue was significantly higher (P = 0.02) for thrombolysed patients than for non-thrombolysed patients, suggesting that thrombolysis rescued tissue with higher infarction risk than typically measured in tissue that spontaneously recovered. The observed spatial heterogeneity of GLM-predicted infarction risk values probably reflects the varying degrees of tissue injury and salvageability that exist after stroke. MRI-based algorithms may therefore provide a more sensitive means for monitoring therapeutic effects on a voxel-wise basis.
    Gottrup C, Thomsen K, Locht P, Wu O, Sorensen GA, Koroshetz WJ, Østergaard L. Applying instance-based techniques to prediction of final outcome in acute stroke. Artif Intell Med 2005;33(3):223-36.Abstract
    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.
    Wu O, Koroshetz WJ, Ostergaard L, Buonanno FS, Copen WA, Gonzalez RG, Rordorf G, Rosen BR, Schwamm LH, Weisskoff RM, Sorensen AG. Predicting tissue outcome in acute human cerebral ischemia using combined diffusion- and perfusion-weighted MR imaging. Stroke 2001;32(4):933-42.Abstract
    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.