Predictive etiological classification of acute ischemic stroke through interpretable machine learning algorithms: a multicenter, prospective cohort study.

BACKGROUND: The prognosis, recurrence rates, and secondary prevention strategies varied significantly among different subtypes of acute ischemic stroke (AIS). Machine learning (ML) techniques can uncover intricate, non-linear relationships within medical data, enabling the identification of factors associated with etiological classification. However, there is currently a lack of research utilizing ML algorithms for predicting AIS etiology. OBJECTIVE: We aimed to use interpretable ML algorithms to develop AIS etiology prediction models, identify critical factors in etiology classification, and enhance existing clinical categorization. METHODS: This study involved patients with the Third China National Stroke Registry (CNSR-III). Nine models, which included Natural Gradient Boosting (NGBoost), Categorical Boosting (CatBoost), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Light Gradient Boosting Machine (LGBM), Gradient Boosting Decision Tree (GBDT), Adaptive Boosting (AdaBoost), Support Vector Machine (SVM), and logistic regression (LR), were employed to predict large artery atherosclerosis (LAA), small vessel occlusion (SVO), and cardioembolism (CE) using an 80:20 randomly split training and test set. We designed an SFS-XGB with 10-fold cross-validation for feature selection. The primary evaluation metrics for the models included the area under the receiver operating characteristic curve (AUC) for discrimination and the Brier score (or calibration plots) for calibration. RESULTS: A total of 5,213 patients were included, comprising 2,471 (47.4%) with LAA, 2,153 (41.3%) with SVO, and 589 (11.3%) with CE. In both LAA and SVO models, the AUC values of the ML models were significantly higher than that of the LR model (P < 0.001). The optimal model for predicting SVO (AUC [RF model] = 0.932) outperformed the optimal LAA model (AUC [NGB model] = 0.917) and the optimal CE model (AUC [LGBM model] = 0.846). Each model displayed relatively satisfactory calibration. Further analysis showed that the optimal CE model could identify potential CE patients in the undetermined etiology (SUE) group, accounting for 1,900 out of 4,156 (45.7%). CONCLUSIONS: The ML algorithm effectively classified patients with LAA, SVO, and CE, demonstrating superior classification performance compared to the LR model. The optimal ML model can identify potential CE patients among SUE patients. These newly identified predictive factors may complement the existing etiological classification system, enabling clinicians to promptly categorize stroke patients' etiology and initiate optimal strategies for secondary prevention.

[Classification of childhood arterial ischemic stroke].

The etiology of childhood arterial ischemic stroke is complex, and identifying the underlying cause is crucial for optimizing treatment and preventing recurrence. Currently, the classification methods for childhood arterial ischemic stroke are largely based on data from international studies, but a unified consensus have not yet been reached. This paper reviews the existing classification methods and their subtype definitions, and points out some doubts and ambiguities. On this basisi, combined with the data collected by Beijing Children's Hospital on Chinese children with arterial ischemic stroke, a new classification method (COIST) was proposed according to the etiology and pathogenesis, namely: inflammation (I), abnormal vascular structure (S), thrombophilia (T), heart disease (C), other identifiable causes (O), and uncertain causes; and various subtypes are listed. It is hoped that this new classification method can attract the attention and discussion of domestic colleagues, with the aim of further refinement, in order to help clinicians better understand and quickly identify the etiologies of childhood ischemic stroke.

Added Value of a Blinded Outcome Adjudication Committee in an Open-Label Randomized Stroke Trial.

BACKGROUND AND PURPOSE: Blinded outcome assessment in trials with prospective randomized open blinded end point design is challenging. Unblinding can result in misclassified outcomes and biased treatment effect estimates. An outcome adjudication committee assures blinded outcome assessment, but the added value for trials with prospective randomized open blinded end point design and subjective outcomes is unknown. We aimed to assess the degree of misclassification of modified Rankin Scale (mRS) scores by a central assessor and its impact on treatment effect estimates in a stroke trial with prospective randomized open blinded end point design. METHODS: We used data from the MR CLEAN (Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands). The primary outcome was the mRS at 90 days. Standardized, algorithm-based telephone interviews to assess the mRS were conducted from a central location by an experienced research nurse, unaware but not formally blinded to treatment allocation (central assessor). Masked reports of these interviews were adjudicated by a blinded outcome committee. Misclassification was defined as an incorrect classification of the mRS by the central assessor. The effect of endovascular treatment on the mRS was assessed with multivariable ordinal logistic regression. RESULTS: In MR CLEAN, 53/500 (10.6%) of the mRS scores were misclassified. The degree and direction of misclassification did not differ between treatment arms (P=0.59). Benefit of endovascular treatment was shown on the mRS when scored by the central assessor (adjusted common odds ratio, 1.60 [95% CI, 1.16-2.21]) and the outcome adjudication committee (adjusted common odds ratio, 1.67 [95% CI, 1.21-2.20]). CONCLUSIONS: Misclassification by the central assessor was small, randomly distributed over treatment arms, and did not affect treatment effect estimates. This study suggests that the added value of a blinded outcome adjudication committee is limited in a stroke trial with prospective randomized open blinded end point design applying standardized, algorithm-based outcome assessment by a central assessor, who is unaware but not formally blinded to treatment allocation. Registration: URL: https://www.isrctn.com; Unique identifier: ISRCTN10888758.

Etiological Subtypes of Transient Ischemic Attack and Ischemic Stroke in Chronic Kidney Disease: Population-Based Study.

BACKGROUND AND PURPOSE: Chronic kidney disease (CKD) is strongly associated with stroke risk, but the mechanisms underlying this association are unclear and might be informed by subtype-specific analyses. However, few studies have reported stroke subtypes in CKD according to established classification systems, such as the TOAST (Trial of ORG 10172 in Acute Stroke Treatment) criteria. We, therefore, aimed to determine which transient ischemic attack and ischemic stroke subtypes using the TOAST classification occur most frequently in patients with CKD. METHODS: In a population-based study of all transient ischemic attack and stroke (OXVASC [Oxford Vascular Study]; 2002-2017), all ischemic events were classified by TOAST subtypes (cardioembolism, large artery disease, small vessel disease, undetermined, multiple, other etiology, or incompletely investigated). Logistic regression was used to determine the relationship between CKD (defined as an estimated glomerular filtration rate <60 mL/min per 1.73 m2) and transient ischemic attack/stroke subtypes adjusted for age, sex, and hypertension and then stratified by age and estimated glomerular filtration rate category. RESULTS: Among 3178 patients with transient ischemic attack (n=1167), ischemic stroke (n=1802), and intracerebral hemorrhage (n=209), 1267 (40%) had CKD. Although there was a greater prevalence of cardioembolic events (31.8% versus 21.2%; P<0.001) in patients with CKD, this association was lost after adjustment for age, sex, and hypertension (adjusted odds ratio=1.20 [95% CI, 0.99-1.45]; P=0.07). Similarly, although patients with CKD had a lower prevalence of small vessel disease (8.8% versus 13.6%; P<0.001), undetermined (26.1% versus 39.4%; P<0.001), and other etiology (1.0% versus 3.6%; P<0.001) subtypes, these associations were also lost after adjustment (adjusted odds ratio=0.86 [0.65-1.13]; P=0.27 and 0.73 [0.36-1.43]; P=0.37 for small vessel disease and other defined etiology, respectively) for all but undetermined (adjusted odds ratio=0.81 [0.67-0.98]; P=0.03). CONCLUSIONS: There were no independent positive associations between CKD and specific TOAST subtypes, which suggest that renal-specific risk factors are unlikely to play an important role in the etiology of particular subtypes. Future studies of stroke and CKD should report subtype-specific analyses to gain further insights into potential mechanisms.

Association of neonatal inflammatory markers and perinatal stroke subtypes.

OBJECTIVE: To examine the relationship between neonatal inflammatory cytokines and perinatal stroke using a systems biology approach analyzing serum and blood-spot cytokines from 47 patients. METHODS: This was a population-based, controlled cohort study with prospective and retrospective case ascertainment. Participants were recruited through the Alberta Perinatal Stroke Project. Stroke was classified as neonatal arterial ischemic stroke (NAIS), arterial presumed perinatal ischemic stroke (APPIS), or periventricular venous infarction (PVI). Biosamples were stored blood spots (retrospective) and acute serum (prospective). Controls had comparable gestational and maternal ages. Sixty-five cytokines were measured (Luminex). Hierarchical clustering analysis was performed to create heat maps. The Fisher linear discriminant analysis was used to create projection models to determine discriminatory boundaries between stroke types and controls. RESULTS: A total of 197 participants were analyzed (27 with NAIS, 8 with APPIS, 12 with PVI, 150 controls). Cytokines were quantifiable with quality control measures satisfied (standards testing, decay analysis). Linear discriminant analysis had high accuracy in using cytokine profiles to separate groups. Profiles in participants with PVI and controls were similar. NAIS separation was accurate (sensitivity 77%, specificity 97%). APPIS mapping was also distinguishable from NAIS (sensitivity 86%, specificity 99%). Classification tree analysis generated similar diagnostic accuracy. CONCLUSIONS: Unique inflammatory biomarker signatures are associated with specific perinatal stroke diseases. Findings support an acquired pathophysiology and suggest the possibility that at-risk pregnancies might be identified to develop prevention strategies. CLASSIFICATION OF EVIDENCE: This study provides Class III evidence that differences in acute neonatal serum cytokine profiles can discriminate between patients with specific perinatal stroke diseases and controls.

Clinical risk factors associated with recurrence of ischemic stroke within two years: A cohort study.

To identify the clinical risk factors and investigate the efficacy of a classification model based on the identified factors for predicting 2-year recurrence after ischemic stroke.From June 2017 to January 2019, 358 patients with first-ever ischemic stroke were enrolled and followed up in Shenzhen Traditional Chinese Medicine Hospital. Demographic and clinical characteristics were recorded by trained medical staff. The outcome was defined as recurrence within 2 years. A multivariate logistic regression model with risk factors and their interaction effects was established and evaluated.The mean (standard deviation) age of the participants was 61.6 (12.1) years, and 101 (28.2%) of the 358 patients were female. The common comorbidities included hypertension (286 patients, 79.9%), diabetes (148 patients, 41.3%), and hyperlipidemia (149 patients, 41.6%). The 2-year recurrence rate was 30.7%. Of the 23 potential risk factors, 10 were significantly different between recurrent and non-recurrent subjects in the univariate analysis. A multivariate logistic regression model was developed based on 10 risk factors. The significant variables include diabetes mellitus, smoking status, peripheral artery disease, hypercoagulable state, depression, 24 h minimum systolic blood pressure, 24 h maximum diastolic blood pressure, age, family history of stroke, NIHSS score status. The area under the receiver operating characteristic curve (ROC) was 0.78 (95% confidence interval: 0.726-0.829) with a sensitivity of 0.61 and a specificity of 0.81, indicating a potential predictive ability.Ten risk factors were identified, and an effective classification model was built. This may aid clinicians in identifying high-risk patients who would benefit most from intensive follow-up and aggressive risk factor reduction.The clinical trial registration number: ChiCTR1800019647.

Serum omentin-1 and risk of one-year mortality in patients with ischemic stroke.

BACKGROUND: Prior studies suggested that ischemic stroke patients with high omentin-1 concentrations were at a decreased risk of unstable carotid plaque and 3-month poor functional outcome. We aim to evaluate the association between serum omentin-1 and 1-y mortality after ischemic stroke. METHODS: A total of 303 ischemic stroke patients were prospectively followed up at 1 y. Outcome was defined as death occurred during the follow-up period. A multivariable Cox model was used to evaluate the association between serum omentin-1 concentrations and 1-y mortality among ischemic stroke patients. RESULTS: From lowest to highest tertile of serum omentin-1, the 1-y cumulative death rate was 12%, 3.7% and 2.1%, respectively (P = 0.006). The hazard ratio (95% confidence interval) of the highest tertile compared with the lowest tertile was 0.19 (0.04-0.88) for mortality after multivariable adjustment (P for trend < 0.01). The net reclassification index and integrated discrimination improvement were significantly improved in predicting 1-y mortality when omentin-1 data was added to the multivariable Cox regression model. CONCLUSIONS: Among patients with ischemic stroke, high baseline serum omentin-1 was associated with a decreased risk of 1-y mortality. These findings, if confirmed by clinical trials, suggest that increasing omentin-1 concentrations may lower the risk of mortality among ischemic stroke patients.

Apolipoproteins B and A1 in Ischemic Stroke Subtypes.

INTRODUCTION: Elevated serum apolipoprotein B and the apolipoprotein B/A1 ratio have been associated with ischemic stroke and intracranial atherosclerotic disease. We sought to assess the relationship between serum levels of apolipoprotein B, apolipoprotein A1, and the apolipoprotein B/A1 ratio with ischemic stroke subtypes and large artery atherosclerosis location. MATERIALS AND METHODS: We evaluated serum apolipoprotein B and apolipoprotein A1 levels in consecutive, statin-naïve, adult ischemic stroke patients admitted to an academic medical center in southern India. We evaluated for differences in the mean serum levels of apolipoprotein B, apolipoprotein A1, and the apolipoprotein B/A1 ratio between patients with ischemic stroke attributed to intracranial atherosclerotic disease, extracranial atherosclerotic disease, small vessel disease, and cardioembolism. In secondary analysis, we assessed for differences in these serum apolipoproteins between patients with moderate-severe intracranial atherosclerotic disease and extracranial atherosclerotic disease, irrespective of ischemic stroke subtype. RESULTS: Among the 156 ischemic stroke patients enrolled in this study, there were no significant differences in serum levels of apolipoprotein B, apolipoprotein A1, and the apolipoprotein B/A1 ratio between patients with distinct ischemic stroke subtypes. No significant differences were found in serum levels of apolipoprotein B, A1 and the apolipoprotein B/A1 ratio between patients with moderate-severe intracranial atherosclerotic disease and moderate-severe extracranial atherosclerotic disease. DISCUSSION: Serum levels of apolipoprotein B and A1 did not differ between ischemic stroke subtypes. Additional studies are needed to validate our findings and to better understand the relationship between serum apolipoproteins and stroke.

Current advances in ischemic stroke research and therapies.

With more than 795,000 cases occurring every year, stroke has become a major problem in the United States across all demographics. Stroke is the leading cause of long-term disability and is the fifth leading cause of death in the US. Ischemic stroke represents 87% of total strokes in the US, and is currently the main focus of stroke research. This literature review examines the risk factors associated with ischemic stroke, changes in cell morphology and signaling in the brain after stroke, and the advantages and disadvantages of in vivo and in vitro ischemic stroke models. Classification systems for stroke etiology are also discussed briefly, as well as current ischemic stroke therapies and new therapeutic strategies that focus on the potential of stem cells to promote stroke recovery.

Reclassification of Ischemic Stroke Etiological Subtypes on the Basis of High-Risk Nonstenosing Carotid Plaque.

Background and Purpose- Carotid artery plaque with <50% luminal stenosis may be an underappreciated stroke mechanism. We assessed how many stroke causes might be reclassified after accounting for nonstenosing plaques with high-risk features. Methods- We included patients enrolled in the Cornell Acute Stroke Academic Registry from 2011 to 2015 who had anterior circulation infarction, magnetic resonance imaging of the brain, and magnetic resonance angiography of the neck. High-risk plaque was identified by intraplaque hemorrhage ascertained from routine neck magnetic resonance angiography studies using validated methods. Infarct location was determined from diffusion-weighted imaging. Intraplaque hemorrhage and infarct location were assessed separately in a blinded fashion by a neuroradiologist. We used the McNemar test for matched data to compare the prevalence of intraplaque hemorrhage ipsilateral versus contralateral to brain infarction. We reclassified stroke subtypes by including large-artery atherosclerosis as a cause if there was intraplaque hemorrhage ipsilateral to brain infarction, regardless of the degree of stenosis. Results- Among the 1721 acute ischemic stroke patients registered in the Cornell Acute Stroke Academic Registry from 2011 to 2015, 579 were eligible for this analysis. High-risk plaque was more common ipsilateral versus contralateral to brain infarction in large-artery atherosclerotic (risk ratio [RR], 3.7 [95% CI, 2.2-6.1]), cryptogenic (RR, 2.1 [95% CI, 1.4-3.1]), and cardioembolic strokes (RR, 1.7 [95% CI, 1.1-2.4]). There were nonsignificant ipsilateral-contralateral differences in high-risk plaque among lacunar strokes (RR, 1.2 [95% CI, 0.4-3.5]) and strokes of other determined cause (RR, 1.5 [95% CI, 0.7-3.3]). After accounting for ipsilateral high-risk plaque, 88 (15.2%) patients were reclassified: 38 (22.6%) cardioembolic to multiple potential etiologies, 6 (8.5%) lacunar to multiple, 3 (15.8%) other determined cause to multiple, and 41 (20.8%) cryptogenic to large-artery atherosclerosis. Conclusions- High-risk carotid plaque was more prevalent ipsilateral to brain infarction across several ischemic stroke subtypes. Accounting for such plaques may reclassify the etiologies of up to 15% of cases in our sample.

Hemodynamic stress distribution identified by SPECT reflects ischemic symptoms of Moyamoya disease patients.

This study aimed to validate the hypothesis that the ratio of cerebral blood flow (CBF) at rest in the lenticular nucleus (LN) territory to that in the middle cerebral artery (MCA) territory is higher in symptomatic Moyamoya disease (MMD) patients than in asymptomatic MMD patients. This was a retrospective observational study of adult patients with documented MMD who underwent single-photon emission computed tomography (SPECT) and had been examined at the Department of Neurosurgery of Keio University Hospital during a 10-year period (2006-2016). The diagnosis was made on the basis of typical imaging findings. We classified unoperated MMD patients into three groups: class I, no evidence of stenosis or occlusion hemispheres and without symptoms in unilateral MMD patients; class II, hemispheres with stenosis or occlusion but without ischemic symptoms; and class III, hemispheres with evidence of stenosis or occlusion associated with ischemic symptoms. Hemodynamic stress distribution (hdSD) was defined as the ratio of CBF in one LN to the CBF in the peripheral MCA; this was obtained by SPECT at rest. We compared the values of CBF and hdSD among the groups. A total of 173 adult patients were diagnosed with MMD from January 1, 2006, to January 1, 2016. Among them, 85 MMD patients underwent SPECT studies. After excluding inappropriate cases, 144 hemispheres were included in our analysis. hdSD was significantly higher (p < 0.001) in hemispheres with ischemic symptoms (class III, mean hdSD = 1.1; 36 sides) than in those without symptoms (class II, mean hdSD = 1.03; 82 sides). However, CBF at rest in the MCA or LN was not significantly associated with ischemic symptoms. The optimal threshold for hdSD to have ischemic symptoms was 1.040 (area under the curve; 74% sensitivity 91.7% and specificity 54.9%). We used SPECT to investigate cerebral blood from MMD patients and found that high hdSD values were predictive of ischemic symptom development in these patients.

Associations Between Features of Nonstenosing Carotid Plaque on Computed Tomographic Angiography and Ischemic Stroke Subtypes.

Background Thromboembolism from nonstenosing carotid plaques may be an underrecognized cause of embolic strokes of undetermined source (ESUS). We evaluated the association between features of nonstenosing atherosclerotic plaque on computed tomographic angiography and ESUS. Methods and Results We identified consecutive acute ischemic stroke patients from 2011 to 2015 who had unilateral anterior territory infarction on brain magnetic resonance imaging and a neck computed tomographic angiography. We included ESUS cases and as controls, cardioembolic strokes. Patients with ≥50% internal carotid artery atherosclerotic stenosis ipsilateral to the stroke were excluded from this analysis. Reviewers blinded to infarct location and stroke cause retrospectively evaluated computed tomographic angiography studies for specific plaque features including thickness of the total, soft, and calcified plaque; presence of ulceration; and perivascular fat attenuation. Paired t tests and McNemar's test for paired data were used to compare plaque features ipsilateral versus contralateral to the side of infarction. Ninety-one patients with ESUS or cardioembolic stroke were included in this study. Total plaque thickness was greater on the infarcted side (2.1±2.0 mm) than the contralateral side (1.2±1.5 mm) (P=0.006) among ESUS cases, but not among cardioembolic cases (1.9±1.6 mm versus 1.8±1.6 mm) (P=0.32). Conclusions Among ESUS cases, total plaque thickness was greater ipsilateral to the side of infarction than on the contralateral, stroke-free side. No such side-to-side differences were apparent in cardioembolic strokes. Our findings suggest that nonstenosing large-artery atherosclerotic plaques represent one underlying mechanism of ESUS.

Causal Effect of Lp(a) [Lipoprotein(a)] Level on Ischemic Stroke and Alzheimer Disease: A Mendelian Randomization Study.

Background and Purpose- Stroke and Alzheimer disease are 2 major causes of neurological disability in aged people and shared overlapping predictors. In recent prospective studies, high Lp(a) [lipoprotein(a)] level is associated with high risk of stroke but low risk of Alzheimer disease. Whether this reflects a causal association remains to be established. The aim of this study is to examine the causal associations of Lp(a) concentrations on ischemic stroke, ischemic stroke subtypes, and Alzheimer disease. Methods- We used 9 single-nucleotide polymorphisms associated with Lp(a) concentrations as instrumental variables. Summary-level data on ischemic stroke and its subtypes were obtained from the Multiancestry Genome-Wide Association Study of Stroke consortium with European individuals ≤446 696 individuals. Summary-level data on Alzheimer disease were obtained from the International Genomics of Alzheimer Project With European individuals ≤54 162 individuals. Two-sample Mendelian randomization (MR) estimates were calculated with inverse-variance weighted, penalized inverse-variance weighted, simple median, weighted median, and MR Pleiotropy Residual Sum and Outlier approaches, and MR-Egger regression was used to explore pleiotropy. Results- Genetically predicted 1-SD log-transformed increase in Lp(a) concentrations was associated with a substantial increase in risk of large artery stroke (odds ratio, 1.20; 95% CI, 1.11-1.30; P<0.001) and a reduce in risk of small vessel stroke (odds ratio, 0.92; 95% CI, 0.88-0.97; P=0.001) and Alzheimer disease (odds ratio, 0.94; 95% CI, 0.91-0.97; P<0.001) using inverse-variance weighted method. No significant association was observed for total ischemic stroke or cardioembolic stroke. MR-Egger indicated no evidence of pleiotropic bias. Results were broadly consistent in sensitivity analyses using penalized inverse-variance weighted, simple median, weighted median, and MR Pleiotropy Residual Sum and Outlier approaches accounting for potential genetic pleiotropy or outliers. Conclusions- This study provides evidence to support that high Lp(a) concentrations was causally associated with an increased risk of large artery stroke but a decreased risk of small vessel stroke and Alzheimer disease. The mechanism underlying the double-edged sword effect of Lp(a) concentrations on neurological system requires further investigation.

Biomarkers for ischemic stroke subtypes: A protein-protein interaction analysis.

According to the Trial of Org 10172 in Acute Stroke Treatment, ischemic stroke is classified into five subtypes. However, the predictive biomarkers of ischemic stroke subtypes are still largely unknown. The utmost objective of this study is to map, construct and analyze protein-protein interaction (PPI) networks for all subtypes of ischemic stroke, and to suggest the predominant biological pathways for each subtypes. Through 6285 protein data retrieved from PolySearch2 and STRING database, the first PPI networks for all subtypes of ischemic stroke were constructed. Notably, F2 and PLG were identified as the critical proteins for large artery atherosclerosis (LAA), lacunar, cardioembolic, stroke of other determined etiology (SOE) and stroke of undetermined etiology (SUE). Gene ontology and DAVID analysis revealed that GO:0030193 regulation of blood coagulation and GO:0051917 regulation of fibrinolysis were the important functional clusters for all the subtypes. In addition, inflammatory pathway was the key etiology for LAA and lacunar, while FOS and JAK2/STAT3 signaling pathways might contribute to cardioembolic stroke. Due to many risk factors associated with SOE and SUE, the precise etiology for these two subtypes remained to be concluded.

Stroke Mechanisms in Symptomatic Intracranial Atherosclerotic Disease: Classification and Clinical Implications.

Background and Purpose- In patients with symptomatic intracranial atherosclerotic stenosis, identifying the underlying stroke mechanisms may inform secondary prevention. We aimed to propose reproducible classification criteria for stroke mechanisms based on routine neuroimaging in symptomatic intracranial atherosclerotic stenosis and explore their clinical implications. Methods- We recruited patients with acute ischemic stroke attributed to 50% to 99% intracranial atherosclerotic stenosis in anterior circulation from 2 centers. Two investigators independently classified probable stroke mechanisms as parent artery atherosclerosis occluding penetrating artery, artery-to-artery embolism, hypoperfusion, and mixed mechanisms, with prespecified criteria based on infarct topography and magnetic resonance/computed tomography angiography. These stroke mechanisms were correlated with features of the patients at baseline and recurrent ischemic stroke in the same territory or relevant transient ischemic attack within 1 year. Results- Among 153 patients recruited, the most common stroke mechanisms were isolated hypoperfusion (35.3%) and mixed mechanism of artery-to-artery embolism and hypoperfusion (37.3%) that was associated with higher incidence of dyslipidemia (P=0.045) and hypertension (P=0.033) than patients with other stroke mechanisms. The proposed criteria showed substantial to excellent intrarater and interrater reproducibilities (κ, 0.791-0.908). Overall, 31 patients received interventional treatment of the diseased intracranial artery; 122 received medical treatment, among whom a mixed mechanism of artery-to-artery embolism and hypoperfusion at baseline was associated with higher risk of ischemic stroke in the same territory within 1 year (24.4% versus 7.8%; hazard ratio, 3.40; 95% CI, 1.25-9.20; log-rank P=0.010) than other mechanisms combined. Conclusions- Artery-to-artery embolism and hypoperfusion commonly coexist in ischemic stroke attributed to intracranial atherosclerotic stenosis, which may be associated with higher risk of stroke relapse.

Comparison of classification methods for tissue outcome after ischaemic stroke.

In acute ischaemic stroke, identifying brain tissue at high risk of infarction is important for clinical decision-making. This tissue may be identified with suitable classification methods from magnetic resonance imaging data. The aim of the present study was to assess and compare the performance of five popular classification methods (adaptive boosting, logistic regression, artificial neural networks, random forest and support vector machine) in identifying tissue at high risk of infarction on human voxel-based brain imaging data. The classification methods were used with eight MRI parameters, including diffusion-weighted imaging and perfusion-weighted imaging obtained in 55 patients. The five criteria used to assess the performance of the methods were the area under the receiver operating curve (AUCroc ), the area under the precision-recall curve (AUCpr ), sensitivity, specificity and the Dice coefficient. The methods performed equally in terms of sensitivity and specificity, while the results of AUCroc and the Dice coefficient were significantly better for adaptive boosting, logistic regression, artificial neural networks and random forest. However, there was no statistically significant difference between the performances of these five classification methods regarding AUCpr , which was the main comparison metric. Machine learning methods can provide valuable prognostic information using multimodal imaging data in acute ischaemic stroke, which in turn can assist in developing personalized treatment decision for clinicians after a thorough validation of methods with an independent data set.

Association of Neutrophil-to-Lymphocyte Ratio and Lymphocyte-to-Monocyte Ratio with Treatment Modalities of Acute Ischaemic Stroke: A Pilot Study.

Background and Objectives: Ischaemic stroke (IS) is the leading cause of death and disability worldwide. All stages of cerebral ischaemia, but especially acute phase, are associated with inflammatory response. Recent studies showed that neutrophil-to-lymphocyte ratio (NLR) and lymphocyte-to-monocyte ratio (LMR) may be used to assess inflammation in IS. To test whether there is a relationship between these parameters and type of stroke treatment, we analysed NLR and LMR in IS patients treated with three different modalities. Materials and Methods: The study included 58 adults with acute IS. A total of 28 patients received intravenous thrombolysis. In another 10 patients, the thrombolytic therapy was followed by thrombectomy and 20 patients did not undergo causal treatment. Blood samples were obtained within 24 h of the stroke diagnosis to calculate NLR and LMR. Next, NLR and LMR of the study subgroups were compared. Results: Our study revealed that NLR was significantly higher in patients treated with thrombectomy following thrombolysis, compared to no causal treatment. Statistical analysis demonstrated that patients with high National Institutes of Health Stroke Scale (NIHSS) scores presented higher NLR than in those with low NIHSS scores. Additionally, patients with high-sensitivity C-reactive protein (hs-CRP) ≥ 3 mg/L presented with significantly higher NLR and significantly lower LMR than the group of patients with lower hs-CRP (<3 mg/L). Conclusions: The main finding of this pilot study was that NLR in IS patients treated using thrombectomy following thrombolysis was markedly higher than that in other treatment groups, which was associated with increased severity of the disease in these patients. Therefore, patients with higher NLR may be expected to have more severe stroke. The link between stroke severity and NLR deserves further study.

Etiological Classification of Cerebral Ischemic Stroke by the TOAST, SSS-TOAST, and ASCOD Systems: The Impact of Observer's Experience on Reliability.

PURPOSE: To assess the impact of observer's experience on reliability of etiological classification systems in patients with ischemic stroke. PATIENTS AND METHODS: We retrospectively reviewed medical records of 80 patients with ischemic stroke in hospitals from August 2016 to March 2017 consecutively. Patients were classified by 4 observers with different clinical experiences and backgrounds (A, B, C, and D) according to the Trial of ORG 10172 in Acute Stroke Treatment (TOAST), Stop Stroke Study TOAST (SSS-TOAST), and ASCOD (A-atherosclerosis, S-small vessel disease, C-cardiac pathology, O-other cause, and D-dissection). The intraobserver reliability was assessed based on the initial and a second delayed assessment after 3 months, and the interobserver reliability of different pairs (A-B and C-D) and overall (A, B, C, and D) were compared based on the initial classification. RESULTS: The reliability values of the 3 classification systems were improved with observer's experience increasing, particularly in the TOAST system, in which the intraobserver reliability values of observers A, B, C, and D were 0.62, 0.73, 0.80, and 0.88, respectively, and slight differences were observed between the SSS-TOAST and ASCOD systems. The A-B pair had lower interobserver reliability value than the C-D pair, particularly in TOAST system with reliability values of 0.36 and 0.74, respectively, and a slight variation of interobserver reliability values were noted in the SSS-TOAST and ASCOD system. CONCLUSIONS: Observer's experience may affect the reliability of etiological classification systems in patients with ischemic stroke.

Automating Ischemic Stroke Subtype Classification Using Machine Learning and Natural Language Processing.

OBJECTIVE: The manual adjudication of disease classification is time-consuming, error-prone, and limits scaling to large datasets. In ischemic stroke (IS), subtype classification is critical for management and outcome prediction. This study sought to use natural language processing of electronic health records (EHR) combined with machine learning methods to automate IS subtyping. METHODS: Among IS patients from an observational registry with TOAST subtyping adjudicated by board-certified vascular neurologists, we analyzed unstructured text-based EHR data including neurology progress notes and neuroradiology reports using natural language processing. We performed several feature selection methods to reduce the high dimensionality of the features and 5-fold cross validation to test generalizability of our methods and minimize overfitting. We used several machine learning methods and calculated the kappa values for agreement between each machine learning approach to manual adjudication. We then performed a blinded testing of the best algorithm against a held-out subset of 50 cases. RESULTS: Compared to manual classification, the best machine-based classification achieved a kappa of .25 using radiology reports alone, .57 using progress notes alone, and .57 using combined data. Kappa values varied by subtype being highest for cardioembolic (.64) and lowest for cryptogenic cases (.47). In the held-out test subset, machine-based classification agreed with rater classification in 40 of 50 cases (kappa .72). CONCLUSIONS: Automated machine learning approaches using textual data from the EHR shows agreement with manual TOAST classification. The automated pipeline, if externally validated, could enable large-scale stroke epidemiology research.