Artificial Intelligence for Clinical Decision Support in Acute Ischemic Stroke: A Systematic Review.

BACKGROUND: Established randomized trial-based parameters for acute ischemic stroke group patients into generic treatment groups, leading to attempts using various artificial intelligence (AI) methods to directly correlate patient characteristics to outcomes and thereby provide decision support to stroke clinicians. We review AI-based clinical decision support systems in the development stage, specifically regarding methodological robustness and constraints for clinical implementation. METHODS: Our systematic review included full-text English language publications proposing a clinical decision support system using AI techniques for direct decision support in acute ischemic stroke cases in adult patients. We (1) describe data and outcomes used in those systems, (2) estimate the systems' benefits compared with traditional stroke diagnosis and treatment, and (3) reported concordance with reporting standards for AI in healthcare. RESULTS: One hundred twenty-one studies met our inclusion criteria. Sixty-five were included for full extraction. In our sample, utilized data sources, methods, and reporting practices were highly heterogeneous. CONCLUSIONS: Our results suggest significant validity threats, dissonance in reporting practices, and challenges to clinical translation. We outline practical recommendations for the successful implementation of AI research in acute ischemic stroke treatment and diagnosis.

Exact Basilar Artery Occlusion Location Indicates Stroke Etiology and Recanalization Success in Patients Eligible for Endovascular Stroke Treatment.

INTRODUCTION: Endovascular stroke treatment (EST) is commonly performed for acute basilar artery occlusion (BAO). We aimed to identify the role of the exact location of BAO in patients receiving EST regarding the stroke etiology, recanalization success and prediction of favorable clinical outcome. METHODS: Retrospective analysis of 191 consecutive patients treated for BAO with EST from 01/2013 until 06/2021 in a tertiary stroke center. Groups were defined according to exact location of BAO in I: proximal third, II: middle third, III: distal third and IV: tip of the basilar artery. Univariate and multivariate analyses were performed for BAO location comparing stroke etiology, recanalization result and favorable clinical outcome according to mRS 0-3 90 days after stroke onset. RESULTS: Occlusion sides types I-IV were evenly distributed (37, 36, 60 and 58 patients). Types I and II were more often associated with large artery atherosclerosis (50 vs. 10 patients, p < 0.001). Distal/tip occlusion (types III/IV) occurred mostly in cardiac embolism or embolic stroke of unknown source (89 vs. 12 in types I/II, p < 0.001). Occlusion site correlated with the underlying stroke etiology (AUC [Area under the curve] 0.89, p < 0.0001, OR [odds ratio] for embolism in type IV: 245). Recanalization rates were higher in patients with distal occlusions (type III/IV OR 3.76, CI [95% confidence interval] 1.51-9.53, p = 0.0076). The BAO site is not predicting favorable clinical outcome. CONCLUSION: The exact basilar artery occlusion site in patients eligible for endovascular stroke treatment reflects the stroke etiology and is associated with differing recanalization success but does not predict favorable clinical outcome.

Comparing Poor and Favorable Outcome Prediction With Machine Learning After Mechanical Thrombectomy in Acute Ischemic Stroke.

Background and Purpose: Outcome prediction after mechanical thrombectomy (MT) in patients with acute ischemic stroke (AIS) and large vessel occlusion (LVO) is commonly performed by focusing on favorable outcome (modified Rankin Scale, mRS 0-2) after 3 months but poor outcome representing severe disability and mortality (mRS 5 and 6) might be of equal importance for clinical decision-making. Methods: We retrospectively analyzed patients with AIS and LVO undergoing MT from 2009 to 2018. Prognostic variables were grouped in baseline clinical (A), MRI-derived variables including mismatch [apparent diffusion coefficient (ADC) and time-to-maximum (Tmax) lesion volume] (B), and variables reflecting speed and extent of reperfusion (C) [modified treatment in cerebral ischemia (mTICI) score and time from onset to mTICI]. Three different scenarios were analyzed: (1) baseline clinical parameters only, (2) baseline clinical and MRI-derived parameters, and (3) all baseline clinical, imaging-derived, and reperfusion-associated parameters. For each scenario, we assessed prediction for favorable and poor outcome with seven different machine learning algorithms. Results: In 210 patients, prediction of favorable outcome was improved after including speed and extent of recanalization [highest area under the curve (AUC) 0.73] compared to using baseline clinical variables only (highest AUC 0.67). Prediction of poor outcome remained stable by using baseline clinical variables only (highest AUC 0.71) and did not improve further by additional variables. Prediction of favorable and poor outcomes was not improved by adding MR-mismatch variables. Most important baseline clinical variables for both outcomes were age, National Institutes of Health Stroke Scale, and premorbid mRS. Conclusions: Our results suggest that a prediction of poor outcome after AIS and MT could be made based on clinical baseline variables only. Speed and extent of MT did improve prediction for a favorable outcome but is not relevant for poor outcome. An MR mismatch with small ischemic core and larger penumbral tissue showed no predictive importance.

Image-to-image generative adversarial networks for synthesizing perfusion parameter maps from DSC-MR images in cerebrovascular disease.

Stroke is a major cause of death or disability. As imaging-based patient stratification improves acute stroke therapy, dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI) is of major interest in image brain perfusion. However, expert-level perfusion maps require a manual or semi-manual post-processing by a medical expert making the procedure time-consuming and less-standardized. Modern machine learning methods such as generative adversarial networks (GANs) have the potential to automate the perfusion map generation on an expert level without manual validation. We propose a modified pix2pix GAN with a temporal component (temp-pix2pix-GAN) that generates perfusion maps in an end-to-end fashion. We train our model on perfusion maps infused with expert knowledge to encode it into the GANs. The performance was trained and evaluated using the structural similarity index measure (SSIM) on two datasets including patients with acute stroke and the steno-occlusive disease. Our temp-pix2pix architecture showed high performance on the acute stroke dataset for all perfusion maps (mean SSIM 0.92-0.99) and good performance on data including patients with the steno-occlusive disease (mean SSIM 0.84-0.99). While clinical validation is still necessary for future studies, our results mark an important step toward automated expert-level perfusion maps and thus fast patient stratification.

Combined Perfusion and Permeability Imaging Reveals Different Pathophysiologic Tissue Responses After Successful Thrombectomy.

Despite successful recanalization of large-vessel occlusions in acute ischemic stroke, individual patients profit to a varying degree. Dynamic susceptibility-weighted perfusion and dynamic T1-weighted contrast-enhanced blood-brain barrier permeability imaging may help to determine secondary stroke injury and predict clinical outcome. We prospectively performed perfusion and permeability imaging in 38 patients within 24 h after successful mechanical thrombectomy of an occlusion of the middle cerebral artery M1 segment. Perfusion alterations were evaluated on cerebral blood flow maps, blood-brain barrier disruption (BBBD) visually and quantitatively on ktrans maps and hemorrhagic transformation on susceptibility-weighted images. Visual BBBD within the DWI lesion corresponded to a median ktrans elevation (IQR) of 0.77 (0.41-1.4) min-1 and was found in all 7 cases of hypoperfusion (100%), in 10 of 16 cases of hyperperfusion (63%), and in only three of 13 cases with unaffected perfusion (23%). BBBD was significantly associated with hemorrhagic transformation (p < 0.001). While BBBD alone was not a predictor of clinical outcome at 3 months (positive predictive value (PPV) = 0.8 [0.56-0.94]), hypoperfusion occurred more often in patients with unfavorable clinical outcome (PPV = 0.43 [0.10-0.82]) compared to hyperperfusion (PPV = 0.93 [0.68-1.0]) or unaffected perfusion (PPV = 1.0 [0.75-1.0]). We show that combined perfusion and permeability imaging reveals distinct infarct signatures after recanalization, indicating the severity of prior ischemic damage. It assists in predicting clinical outcome and may identify patients at risk of stroke progression.

Diagnostic and prognostic benefit of arterial spin labeling in subacute stroke.

BACKGROUND AND PURPOSE: Brain perfusion measurement in the subacute phase of stroke may support therapeutic decisions. We evaluated whether arterial spin labeling (ASL), a noninvasive perfusion imaging technique based on magnetic resonance imaging (MRI), adds diagnostic and prognostic benefit to diffusion-weighted imaging (DWI) in subacute stroke. METHODS: In a single-center imaging study, patients with DWI lesion(s) in the middle cerebral artery (MCA) territory were included. Onset to imaging time was ≤7 days and imaging included ASL and DWI sequences. Qualitative (standardized visual analysis) and quantitative perfusion analyses (region of interest analysis) were performed. Dichotomized early outcome (modified Rankin Scale [mRS] 0-2 vs. 3-6) was analyzed in two logistic regression models. Model 1 included DWI lesion volume, age, vascular pathology, admission NIHSS, and acute stroke treatment as covariates. Model 2 added the ASL-based perfusion pattern to Model 1. Receiver-operating-characteristic (ROC) and area-under-the-curve (AUC) were calculated for both models to assess their predictive power. The likelihood-ratio-test compared both models. RESULTS: Thirty-eight patients were included (median age 70 years, admission NIHSS 4, onset to imaging time 67 hr, discharge mRS 2). Qualitative perfusion analysis yielded additional diagnostic information in 84% of the patients. In the quantitative analysis, AUC for outcome prediction was 0.88 (95% CI 0.77-0.99) for Model 1 and 0.97 (95% CI 0.91-1.00) for Model 2. Inclusion of perfusion data significantly improved performance and outcome prediction (p = 0.002) of stroke imaging. CONCLUSIONS: In patients with subacute stroke, our study showed that adding perfusion imaging to structural imaging and clinical data significantly improved outcome prediction. This highlights the usefulness of ASL and noninvasive perfusion biomarkers in stroke diagnosis and management.

Correction for Susceptibility Distortions Increases the Performance of Arterial Spin Labeling in Patients with Cerebrovascular Disease.

BACKGROUND AND PURPOSE: Arterial spin labeling (ASL) is an MRI technique to measure cerebral blood flow (CBF) without the need of exogenous contrast agents and is thus a promising alternative to the clinical standard dynamic susceptibility-weighted contrast-enhanced (DSC) perfusion imaging. Latest international guidelines encourage its application in the clinical setting. However, susceptibility-induced image distortions impair ASL with fast readout modules (eg Echo Planar Imaging, EPI; gradient and spin echo, GRASE). In the present study, we investigated the benefit of a distortion correction for ASL compared to DSC. METHODS: A pulsed ASL (PASL) sequence combined with a 3D-GRASE readout at multiple inflow times (multi-TI) was used and was corrected for susceptibility distortions using a FMRIB Software Library (FSL) implemented tool TOPUP. We performed qualitative (three expert raters) and quantitative (volume of interest [VOI]-based) comparisons of ASL and DSC imaging in 13 patients with chronic steno-occlusive disease. RESULTS: In the qualitative analysis, distortion correction of the images led to a strong increase in diagnostic precision of ASL compared to DSC in the anterior cerebral artery (ACA) perfusion territory, where the susceptibility artifact was most pronounced (specificity 8% vs. 75%). In the quantitative analysis, the correlation between ASL and DSC values increased for all perfusion territories with the best improvement for the ACA territory (for anterior, middle and posterior cerebral artery: ACA: rho -0.22 vs. 0.71; MCA: rho 0.58 vs. 0.76; PCA: rho 0.58 vs. 0.63). CONCLUSIONS: We showed that susceptibility distortion correction strongly improves the comparability of multi-TI ASL 3D-GRASE to DSC in steno-occlusive disease. We suggest it to be implemented in ASL postprocessing routines.

3D GRASE pulsed arterial spin labeling at multiple inflow times in patients with long arterial transit times: comparison with dynamic susceptibility-weighted contrast-enhanced MRI at 3 Tesla.

Pulsed arterial spin labeling (PASL) at multiple inflow times (multi-TIs) is advantageous for the measurement of brain perfusion in patients with long arterial transit times (ATTs) as in steno-occlusive disease, because bolus-arrival-time can be measured and blood flow measurements can be corrected accordingly. Owing to its increased signal-to-noise ratio, a combination with a three-dimensional gradient and spin echo (GRASE) readout allows acquiring a sufficient number of multi-TIs within a clinically feasible acquisition time of 5 minutes. We compared this technique with the clinical standard dynamic susceptibility-weighted contrast-enhanced imaging-magnetic resonance imaging in patients with unilateral stenosis >70% of the internal carotid or middle cerebral artery (MCA) at 3 Tesla. We performed qualitative (assessment by three expert raters) and quantitative (region of interest (ROI)/volume of interest (VOI) based) comparisons. In 43 patients, multi-TI PASL-GRASE showed perfusion alterations with moderate accuracy in the qualitative analysis. Quantitatively, moderate correlation coefficients were found for the MCA territory (ROI based: r=0.52, VOI based: r=0.48). In the anterior cerebral artery (ACA) territory, a readout related right-sided susceptibility artifact impaired correlation (ROI based: r=0.29, VOI based: r=0.34). Arterial transit delay artifacts were found only in 12% of patients. In conclusion, multi-TI PASL-GRASE can correct for arterial transit delay in patients with long ATTs. These results are promising for the transfer of ASL to the clinical practice.

DWI intensity values predict FLAIR lesions in acute ischemic stroke.

BACKGROUND AND PURPOSE: In acute stroke, the DWI-FLAIR mismatch allows for the allocation of patients to the thrombolysis window (<4.5 hours). FLAIR-lesions, however, may be challenging to assess. In comparison, DWI may be a useful bio-marker owing to high lesion contrast. We investigated the performance of a relative DWI signal intensity (rSI) threshold to predict the presence of FLAIR-lesions in acute stroke and analyzed its association with time-from-stroke-onset. METHODS: In a retrospective, dual-center MR-imaging study we included patients with acute stroke and time-from-stroke-onset ≤12 hours (group A: n = 49, 1.5T; group B: n = 48, 3T). DW- and FLAIR-images were coregistered. The largest lesion extent in DWI defined the slice for further analysis. FLAIR-lesions were identified by 3 raters, delineated as regions-of-interest (ROIs) and copied on the DW-images. Circular ROIs were placed within the DWI-lesion and labeled according to the FLAIR-pattern (FLAIR+ or FLAIR-). ROI-values were normalized to the unaffected hemisphere. Adjusted and nonadjusted receiver-operating-characteristics (ROC) curve analysis on patient level was performed to analyze the ability of a DWI- and ADC-rSI threshold to predict the presence of FLAIR-lesions. Spearman correlation and adjusted linear regression analysis was performed to assess the relationship between DWI-intensity and time-from-stroke-onset. RESULTS: DWI-rSI performed well in predicting lesions in FLAIR-imaging (mean area under the curve (AUC): group A: 0.84; group B: 0.85). An optimal mean DWI-rSI threshold was identified (A: 162%; B: 161%). ADC-maps performed worse (mean AUC: A: 0.58; B: 0.77). Adjusted regression models confirmed the superior performance of DWI-rSI. Correlation coefficents and linear regression showed a good association with time-from-stroke-onset for DWI-rSI, but not for ADC-rSI. CONCLUSION: An easily assessable DWI-rSI threshold identifies the presence of lesions in FLAIR-imaging with good accuracy and is associated with time-from-stroke-onset in acute stroke. This finding underlines the potential of a DWI-rSI threshold as a marker of lesion age.

Clinical evaluation of an arterial-spin-labeling product sequence in steno-occlusive disease of the brain.

INTRODUCTION: In brain perfusion imaging, arterial spin labeling (ASL) is a noninvasive alternative to dynamic susceptibility contrast-magnetic resonance imaging (DSC-MRI). For clinical imaging, only product sequences can be used. We therefore analyzed the performance of a product sequence (PICORE-PASL) included in an MRI software-package compared with DSC-MRI in patients with steno-occlusion of the MCA or ICA >70%. METHODS: Images were acquired on a 3T MRI system and qualitatively analyzed by 3 raters. For a quantitative analysis, cortical ROIs were placed in co-registered ASL and DSC images. Pooled data for ASL-cerebral blood flow (CBF) and DSC-CBF were analyzed by Spearman's correlation and the Bland-Altman (BA)-plot. RESULTS: In 28 patients, 11 ASL studies were uninterpretable due to patient motion. Of the remaining patients, 71% showed signs of delayed tracer arrival. A weak correlation for DSC-relCBF vs ASL-relCBF (r = 0.24) and a large spread of values in the BA-plot owing to unreliable CBF-measurement was found. CONCLUSION: The PICORE ASL product sequence is sensitive for estimation of delayed tracer arrival, but cannot be recommended to measure CBF in steno-occlusive disease. ASL-sequences that are less sensitive to patient motion and correcting for delayed blood flow should be available in the clinical setting.

Ultrahigh-field MRI in human ischemic stroke--a 7 tesla study.

INTRODUCTION: Magnetic resonance imaging (MRI) using field strengths up to 3 Tesla (T) has proven to be a powerful tool for stroke diagnosis. Recently, ultrahigh-field (UHF) MRI at 7 T has shown relevant diagnostic benefits in imaging of neurological diseases, but its value for stroke imaging has not been investigated yet. We present the first evaluation of a clinically feasible stroke imaging protocol at 7 T. For comparison an established stroke imaging protocol was applied at 3 T. METHODS: In a prospective imaging study seven patients with subacute and chronic stroke were included. Imaging at 3 T was immediately followed by 7 T imaging. Both protocols included T1-weighted 3D Magnetization-Prepared Rapid-Acquired Gradient-Echo (3D-MPRAGE), T2-weighted 2D Fluid Attenuated Inversion Recovery (2D-FLAIR), T2-weighted 2D Fluid Attenuated Inversion Recovery (2D-T2-TSE), T2* weighted 2D Fast Low Angle Shot Gradient Echo (2D-HemoFLASH) and 3D Time-of-Flight angiography (3D-TOF). RESULTS: The diagnostic information relevant for clinical stroke imaging obtained at 3 T was equally available at 7 T. Higher spatial resolution at 7 T revealed more anatomical details precisely depicting ischemic lesions and periinfarct alterations. A clear benefit in anatomical resolution was also demonstrated for vessel imaging at 7 T. RF power deposition constraints induced scan time prolongation and reduced brain coverage for 2D-FLAIR, 2D-T2-TSE and 3D-TOF at 7 T versus 3 T. CONCLUSIONS: The potential of 7 T MRI for human stroke imaging is shown. Our pilot study encourages a further evaluation of the diagnostic benefit of stroke imaging at 7 T in a larger study.