Ensemble of Convolutional Neural Networks Improves Automated Segmentation of Acute Ischemic Lesions Using Multiparametric Diffusion-Weighted MRI.

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 cm3). 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 (P < .001). Automated volumes correlated with manually measured volumes (Spearman ρ = 0.91, P < .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; P < .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.

Combining MRI with NIHSS thresholds to predict outcome in acute ischemic stroke: value for patient selection.

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.

Exposing hidden truncation-related errors in acute stroke perfusion imaging.

BACKGROUND AND PURPOSE: The durations of acute ischemic stroke patients' CT or MR perfusion scans may be too short to fully sample the passage of the injected contrast agent through the brain. We tested the potential magnitude of hidden errors related to the truncation of data by short perfusion scans. MATERIALS AND METHODS: Fifty-seven patients with acute ischemic stroke underwent perfusion MR imaging within 12 hours of symptom onset, using a relatively long scan duration (110 seconds). Shorter scan durations (39.5-108.5 seconds) were simulated by progressively deleting the last-acquired images. CBV, CBF, MTT, and time to response function maximum (Tmax) were measured within DWI-identified acute infarcts, with commonly used postprocessing algorithms. All measurements except Tmax were normalized by dividing by the contralateral hemisphere values. The effects of the scan duration on these hemodynamic measurements and on the volumes of lesions with Tmax of >6 seconds were tested using regression. RESULTS: Decreasing scan duration from 110 seconds to 40 seconds falsely reduced perfusion estimates by 47.6%-64.2% of normal for CBV, 1.96%-4.10% for CBF, 133%-205% for MTT, and 6.2-8.0 seconds for Tmax, depending on the postprocessing method. This truncation falsely reduced estimated Tmax lesion volume by 71.5 or 93.8 mL, depending on the deconvolution method. "Lesion reversal" (ie, change from above-normal to apparently normal, or from >6 seconds to ≤6 seconds for the time to response function maximum) with increasing truncation occurred in 37%-46% of lesions for CBV, 2%-4% for CBF, 28%-54% for MTT, and 42%-44% for Tmax, depending on the postprocessing method. CONCLUSIONS: Hidden truncation-related errors in perfusion images may be large enough to alter patient management or affect outcomes of clinical trials.

Cerebral blood flow thresholds for tissue infarction in patients with acute ischemic stroke treated with intra-arterial revascularization therapy depend on timing of reperfusion.

BACKGROUND AND PURPOSE: MR perfusion CBF values can distinguish hypoperfused penumbral tissue likely to infarct from that which is likely to recover. Our aim was to determine if CBF thresholds for tissue infarction depend on the timing of recanalization in patients with acute stroke treated with IAT. MATERIALS AND METHODS: Twenty-six patients with acute proximal anterior circulation strokes underwent DWI and PWI before IAT. rCBF was obtained in the following areas: 1) C with abnormal DWI, reduced CBF, follow-up infarction; 2) PI with normal DWI, reduced CBF, follow-up infarction and 3) PNI with normal DWI, reduced CBF, normal follow-up. rCBF in tissue reperfused at <6 hours (early recanalizers), in tissue reperfused at >6 hours (late RC), and in NRC was compared. RESULTS: For C, mean rCBF was 0.13 (SEM, 0.002), 0.29 (0.007), and 0.21 (0.004) for early recanalizers, late recanalizers, and nonrecanalizers, respectively (P < .001, for all comparisons). For PI, mean rCBF was 0.34 (0.006), 0.38 (0.008), and 0.39 (0.005) for early recanalizers, late recanalizers, and nonrecanalizers, respectively (P < .001 for early-versus-late recanalizers and versus nonrecanalizers; P > .05 for late recanalizers versus nonrecanalizers). For PNI, the mean rCBF was 0.38 (0.002), 0.48 (0.003), and 0.48 (0.004) for early recanalizers, late recanalizers, and nonrecanalizers, respectively (P < .001 for early-versus-late recanalizers and nonrecanalizers; P > .05 for late recanalizers versus nonrecanalizers). ROC analyzis demonstrated optimal rCBF thresholds for tissue infarction of 0.27 (sensitivity, 80%; specificity, 87%), 0.44 (sensitivity, 77%; specificity, 75%), and 0.41 (sensitivity, 78%; specificity, 77%) for early recanalizers, late recanalizers, and nonrecanalizers, respectively. CONCLUSIONS: CBF thresholds for tissue infarction in patients with acute stroke are lower in tissue that is reperfused at earlier time points. This information may be important in selecting patients who might benefit from reperfusion therapy.

Ischemic stroke: effects of etiology and patient age on the time course of the core apparent diffusion coefficient.

PURPOSE: To determine whether the evolution of the core apparent diffusion coefficient (ADC) of water in ischemic stroke varies with patient age or infarct etiology. MATERIALS AND METHODS: One hundred forty-seven patients with stroke underwent 236 diffusion-weighted magnetic resonance imaging examinations. Etiologies of lesions were classified according to predefined criteria; in 224 images, the diagnosis of lacune could be firmly established or excluded. ADC was measured in the center of each lesion and in contralateral normal-appearing brain. A model was used to describe the time course of relative ADC (rADC), which is calculated by dividing the lesion ADC by the contralateral ADC, and to test for age- or etiology-related differences in this time course. RESULTS: Transition from decreasing to increasing rADC was estimated at 18.5 hours after stroke onset. In subgroup analysis, transition was earlier in nonlacunes than in lacunes (P =.02). There was a trend toward earlier transition in patients older than the median age of 66.0 years, compared with younger patients (P =.06). Pseudonormalization was estimated at 216 hours. Among nonlacunes, the rate of subsequent rADC increase was more rapid in younger patients than in older patients (P =.001). Within the smaller sample of lacunes, however, no significant age-related difference in this rate was found. CONCLUSION: Differences in ADC depending on the patient's age and infarct etiology suggest differing rates of ADC progression.

Predicting tissue outcome in acute human cerebral ischemia using combined diffusion- and perfusion-weighted MR imaging.

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.

Human acute cerebral ischemia: detection of changes in water diffusion anisotropy by using MR imaging.

PURPOSE: To (a) determine the optimal choice of a scalar metric of anisotropy and (b) determine by means of magnetic resonance imaging if changes in diffusion anisotropy occurred in acute human ischemic stroke. MATERIALS AND METHODS: The full diffusion tensor over the entire brain was measured. To optimize the choice of a scalar anisotropy metric, the performances of scalar indices in simulated models and in a healthy volunteer were analyzed. The anisotropy, trace apparent diffusion coefficient (ADC), and eigenvalues of the diffusion tensor in lesions and contralateral normal brain were compared in 50 patients with stroke. RESULTS: Changes in anisotropy in patients were quantified by using fractional anisotropy because it provided the best performance in terms of contrast-to-noise ratio as a function of signal-to-noise ratio in simulations. The anisotropy of ischemic white matter decreased (P = .01). Changes in anisotropy in ischemic gray matter were not significant (P = .63). The trace ADC decreased for ischemic gray matter and white matter (P < .001). The first and second eigenvalues decreased in both ischemic gray and ischemic white matter (P < .001). The third eigenvalue decreased in ischemic gray (P = .001) and white matter (P = .03). CONCLUSION: Gray matter is mildly anisotropic in normal and early ischemic states. However, early white matter ischemia is associated with not only changes in trace ADC values but also significant changes in the anisotropy, or shape, of the water self-diffusion tensor.

Hyperacute stroke: simultaneous measurement of relative cerebral blood volume, relative cerebral blood flow, and mean tissue transit time.

PURPOSE: To investigate additional information provided by maps of relative cerebral blood flow in functional magnetic resonance (MR) imaging of human hyperacute cerebral ischemic stroke. MATERIALS AND METHODS: Diffusion-weighted and hemodynamic MR imaging were performed in 23 patients less than 12 hours after the onset of symptoms. Maps of relative cerebral blood flow and tracer mean tissue transit time were computed, as were maps of apparent diffusion and relative cerebral blood volume. Acute lesion volumes on the maps were compared with follow-up imaging findings. RESULTS: In 15 of 23 subjects (65%), blood flow maps revealed hemodynamic abnormalities not visible on blood volume maps. A mismatch between initial blood flow and diffusion findings predicted growth of infarct more often (12 of 15 subjects with infarcts that grew) than did a mismatch between initial blood volume and diffusion findings (eight of 15). However, lesion volumes on blood volume and diffusion maps correlated better with eventual infarct volumes (r > 0.90) than did those on blood flow and tracer mean transit time maps (r approximately 0.6), likely as a result of threshold effects. In eight patients, blood volume was elevated around the diffusion abnormality, suggesting a compensatory hemodynamic response. CONCLUSION: MR imaging can delineate areas of altered blood flow, blood volume, and water mobility in hyperacute human stroke. Predictive models of tissue outcome may benefit by including computation of both relative cerebral blood flow and blood volume.

Diffusion- and perfusion-weighted imaging in vasospasm after subarachnoid hemorrhage.

BACKGROUND AND PURPOSE: Better measures of cerebral tissue perfusion and earlier detection of ischemic injury are needed to guide therapy in subarachnoid hemorrhage (SAH) patients with vasospasm. We sought to identify tissue ischemia and early ischemic injury with combined diffusion-weighted (DW) and hemodynamically weighted (HW) MRI in patients with vasospasm after SAH. METHODS: Combined DW and HW imaging was used to study 6 patients with clinical and angiographic vasospasm, 1 patient without clinical signs of vasospasm but with severe angiographic vasospasm, and 1 patient without angiographic spasm. Analysis of the passage of an intravenous contrast bolus through brain was used to construct multislice maps of relative cerebral blood volume (rCBV), relative cerebral blood flow (rCBF), and tissue mean transit time (tMTT). We hypothesize that large HW imaging (HWI) abnormalities would be present in treated patients at the time they develop neurological deficit due to vasospasm without matching DW imaging (DWI) abnormalities. RESULTS: Small, sometimes multiple, ischemic lesions on DWI were seen encircled by a large area of decreased rCBF and increased tMTT in all patients with symptomatic vasospasm. Decreases in rCBV were not prominent. MRI hemodynamic abnormalities occurred in regions supplied by vessels with angiographic vasospasm or in their watershed territories. All patients with neurological deficit showed an area of abnormal tMTT much larger than the area of DWI abnormality. MRI images were normal in the asymptomatic patient with angiographic vasospasm and the patient with normal angiogram and no clinical signs of vasospasm. CONCLUSIONS: We conclude that DW/HW MRI in symptomatic vasospasm can detect widespread changes in tissue hemodynamics that encircle early foci of ischemic injury. With additional study, the technique could become a useful tool in the clinical management of patients with SAH.

Time course of lesion development in patients with acute stroke: serial diffusion- and hemodynamic-weighted magnetic resonance imaging.

BACKGROUND AND PURPOSE: We sought to characterize the evolution of acute ischemic stroke by MRI and its relationship to patients' neurological outcome. METHODS: Fourteen patients with acute ischemic stroke underwent MRI within 13 hours of symptom onset (mean, 7.4+/-3 hours) and underwent repeated imaging and concurrent neurological examination at 8, 24, 36, and 48 hours and 7 days and >42 days after first imaging. RESULTS: Diffusion-weighted imaging (DWI) lesion volumes increased between the first and second scans in 10 of 14 patients; scans with maximum DWI lesion volume occurred at a mean of 70.4 hours. Initial DWI lesion volume correlated with the largest T2 lesion volume (r=0.97; P<0.001). Final lesion volume was smaller than maximum lesion volume in 12 of 14 patients. There was positive correlation between the follow-up National Institutes of Health Stroke Scale score and the initial DWI lesion volume (r=0.67; P=0. 01) and maximum T2 lesion volume (r=0.77; P<0.01) and negative correlation with initial mean apparent diffusion coefficient ratio (ADCr) (r=-0.64; P<0.05). The ADCr was 0.73 at initial imaging and fell between the initial and second scans in 10 of 14 patients. Mean ADCr did not rise above normal until 42 days after stroke onset (P<0. 001). CONCLUSIONS: Serial MRI demonstrates the dynamic nature of progressive ischemic injury in acute stroke patients developing over hours to days, and it suggests that both primary and secondary pathophysiological processes can be valuable targets for neuroprotective interventions.

Regional ischemia and ischemic injury in patients with acute middle cerebral artery stroke as defined by early diffusion-weighted and perfusion-weighted MRI.

BACKGROUND AND PURPOSE: We sought to map early regional ischemia and infarction in patients with middle cerebral artery (MCA) stroke and compare them with final infarct size using advanced MRI techniques. MRI can now delineate very early infarction by diffusion-weighted imaging (DWI) and abnormal tissue perfusion by perfusion-weighted imaging (PWI). METHODS: Seventeen patients seen within 12 hours of onset of MCA stroke had MR angiography, standard MRI, and PWI and DWI MRI. PWI maps were generated by analysis of the passage of intravenous contrast bolus through the brain. Cerebral blood volume (CBV) was determined after quantitative analysis of PWI data. Volumes of the initial DWI and PWI lesion were calculated and compared with a final infarct volume from a follow-up imaging study (CT scan or MRI). RESULTS: Group 1 (10 patients) had MCA stem (M1) occlusion by MR angiography. DWI lesion volumes were smaller than the volumes of CBV abnormality. In 7 patients the final stroke volume was larger or the same, and in 3 it was smaller than the initial CBV lesion. Group 2 (7 patients) had an open M1 on MR angiography with distal MCA stroke. In 6 group 2 patients, the initial DWI lesion matched the initial CBV abnormality and the final infarct. CONCLUSIONS: Most patients with M1 occlusion showed progression of infarction into the region of abnormal perfusion. In contrast, patients with open M1 had strokes consistent with distal branch occlusion and had maximal extent of injury on DWI at initial presentation. Application of these MRI techniques should improve definition of different acute stroke syndromes and facilitate clinical decision making.

Hyperacute stroke: evaluation with combined multisection diffusion-weighted and hemodynamically weighted echo-planar MR imaging.

PURPOSE: To evaluate acute stroke with conventional, multisection diffusion-weighted (DW), and hemodynamically weighted (HW) magnetic resonance (MR) imaging. MATERIALS AND METHODS: The three MR imaging techniques were performed in 11 patients within 10 hours of the onset of acute hemiparesis. The volume of DW and HW abnormalities were compared with infarct volumes depicted at initial and/or follow-up MR or computed tomography (CT). RESULTS: Findings at DW and HW imaging were abnormal in nine of the 11 patients, despite normal findings at initial CT and/or MR. In all nine patients, infarcts were depicted at follow-up CT or MR. The DW abnormality was generally smaller and the HW abnormality was generally larger than the infarct volume determined at subsequent imaging. In the two patients with normal findings at DW and HW imaging, symptoms resolved completely within 1-48 hours. CONCLUSION: Different aspects of hyperacute cerebral ischemia are depicted at DW and HW imaging before infarction is depicted at conventional MR or CT. These techniques may improve stroke diagnosis and may contribute to advances in treatment.