Use of Gradient Boosting Machine Learning to Predict Patient Outcome in Acute Ischemic Stroke on the Basis of Imaging, Demographic, and Clinical Information.
AJR. American journal of roentgenology
OBJECTIVE: When treatment decisions are being made for patients with acute ischemic stroke, timely and accurate outcome prediction plays an important role. The optimal rehabilitation strategy also relies on long-term outcome predictions. The decision-making process involves numerous biomarkers including imaging features and demographic information. The objective of this study was to integrate common stroke biomarkers using machine learning methods and predict patient recovery outcome at 90 days.MATERIALS AND METHODS: A total of 512 patients were enrolled in this retrospective study. Extreme gradient boosting (XGB) and gradient boosting machine (GBM) models were used to predict modified Rankin scale (mRS) scores at 90 days using biomarkers available at admission and 24 hours. Feature selections were performed using a greedy algorithm. Fivefold cross validation was applied to estimate model performance.RESULTS: For binary prediction of an mRS score of greater than 2 using biomarkers available at admission, XGB and GBM had an AUC of 0.746 and 0.748, respectively. Adding the National Institutes of Health Stroke Score at 24 hours and performing feature selection improved the AUC of XGB to 0.884 and the AUC of GBM to 0.877. With the addition of the recanalization outcome, XGB's AUC improved to 0.807 for nonrecanalized patients and dropped to 0.670 for recanalized patients. GBM's AUC improved to 0.781 for nonrecanalized patients and dropped to 0.655 for recanalized patients.CONCLUSION: Decision tree-based GBMs can predict the recovery outcome of stroke patients at admission with a high AUC. Breaking down the patient groups on the basis of recanalization and nonrecanalization can potentially help with the treatment decision process.
View details for PubMedID 30354266
- Optimal Delay Time of CT Perfusion for Predicting Cerebral Parenchymal Hematoma After Intra-Arterial tPA Treatment FRONTIERS IN NEUROLOGY 2018; 9
Reduced Intravoxel Incoherent Motion Microvascular Perfusion Predicts Delayed Cerebral Ischemia and Vasospasm After Aneurysm Rupture.
Proximal artery vasospasm and delayed cerebral ischemia (DCI) after cerebral aneurysm rupture result in reduced cerebral perfusion and microperfusion and significant morbidity and mortality. Intravoxel incoherent motion (IVIM) magnetic resonance imaging extracts microvascular perfusion information from a multi-b value diffusion-weighted sequence. We determined whether decreased IVIM perfusion may identify patients with proximal artery vasospasm and DCI.We performed a pilot retrospective cohort study of patients with ruptured cerebral aneurysms. Consecutive patients who underwent a brain magnetic resonance imaging with IVIM after ruptured aneurysm treatment were included. Patient demographic, treatment, imaging, and outcome data were determined by electronic medical record review. Primary outcome was DCI development with proximal artery vasospasm that required endovascular treatment. Secondary outcomes included mortality and clinical outcomes at 6 months.Sixteen patients (11 females, 69%;P=0.9) were included. There were no differences in age, neurological status, or comorbidities between patients who subsequently underwent endovascular treatment of DCI (10 patients; DCI+ group) and those who did not (6 patients; DCI- group). Compared with DCI- patients, DCI+ patients had decreased IVIM perfusion fractionf(0.09±0.03 versus 0.13±0.01;P=0.03), reduced diffusion coefficientD(0.82±0.05 versus 0.92±0.07×10-3mm2/s;P=0.003), and reduced blood flow-related parameterfD* (1.18±0.40 versus 1.83±0.40×10-3mm2/s;P=0.009). IVIM pseudodiffusion coefficientD* did not differ between DCI- (0.011±0.002) and DCI+ (0.013±0.005 mm2/s;P=0.4) patients. No differences in mortality or clinical outcome were identified.Decreased IVIM perfusion fractionfand blood flow-related parameterfD* correlate with DCI and proximal artery vasospasm development after cerebral aneurysm rupture.
View details for DOI 10.1161/STROKEAHA.117.020395
View details for PubMedID 29439196
Assessing the Relationship between Atherosclerotic Cardiovascular Disease Risk Score and Carotid Artery Imaging Findings.
Journal of neuroimaging : official journal of the American Society of Neuroimaging
To characterize the relationship between computed tomography angiography (CTA) imaging characteristics of carotid artery and the 10-year risk of atherosclerotic cardiovascular disease (ASCVD) score.We retrospectively identified all patients who underwent a cervical CTA at our institution from January 2013 to July 2016, extracted clinical information, and calculated the 10-year ASCVD score using the Pooled Cohort Equations from the 2013 ACC/AHA guidelines. We compared the imaging features of artery atherosclerosis derived from the CTAs between low and high risk.One hundred forty-six patients met our inclusion criteria. Patients with an ASCVD score ≥7.5% (64.4%) had significantly more arterial stenosis than patients with an ASCVD score <7.5% (35.6%, P < .001). Maximal plaque thickness was significantly higher (mean 2.33 vs. .42 mm, P < .001) and soft plaques (55.3% vs. 13.5%, P < .001) were significantly more frequent in patients with an ASCVD score ≥7.5%. However, among patients with a 10-year ASCVD score ≥7.5%, 33 (35.1%) had no arterial stenosis, 35 (37.2%) had a maximal plaque thickness less than. 9 mm, and 42 (44.7%) had no soft plaque. Furthermore, among the patients with a 10-year ASCVD score <7.5%, 8 (15.4%) had some arterial stenosis, 8 (15.4%) had a maximal plaque thickness more than. 9 mm, and 7 (13.5%) had soft plaque.There is some concordance but not a perfect overlap between the 10-year ASCVD risk scores calculated from clinical and blood assessment and carotid artery imaging findings.
View details for PubMedID 30357980
Assessing the Relationship Between American Heart Association Atherosclerotic Cardiovascular Disease Risk Score and Coronary Artery Imaging Findings.
Journal of computer assisted tomography
The aim of this study was to characterize the relationship between computed tomography angiography imaging characteristics of coronary artery and atherosclerotic cardiovascular disease (ASCVD) score.We retrospectively identified all patients who underwent a coronary computed tomography angiography at our institution from December 2013 to July 2016, then we calculated the 10-year ASCVD score. We characterized the relationship between coronary artery imaging findings and ASCVD risk score.One hundred fifty-one patients met our inclusion criteria. Patients with a 10-year ASCVD score of 7.5% or greater had significantly more arterial segments showing stenosis (46.4%, P = 0.008) and significantly higher maximal plaque thickness (1.25 vs 0.53, P = 0.001). However, among 56 patients with a 10-year ASCVD score of 7.5% or greater, 30 (53.6%) had no arterial stenosis. Furthermore, among the patients with a 10-year ASCVD score of less than 7.5%, 24 (25.3%) had some arterial stenosis.There is some concordance but not a perfect overlap between 10-year ASCVD risk scores and coronary artery imaging findings.
View details for PubMedID 30407249
CT Permeability Imaging Predicts Clinical Outcomes in Acute Ischemic Stroke Patients Treated with Intra-arterial Thrombolytic Therapy
2017; 54 (4): 2539-2546
In this study, we determined whether a prediction of final infarct volume (FIV) and clinical outcomes in patients with an acute stroke is improved by using a contrast transfer coefficient (K (trans)) as a biomarker for blood-brain barrier (BBB) dysfunction. Here, consecutive patients admitted with signs and symptoms suggesting acute hemispheric stroke were included in this study. Ninety-eight participants with intra-arterial therapy were assessed (46 female). Definition of predicted FIV was performed using conventional perfusion CT (PCT-PIV) parameters alone and in combination with K (trans) (K (trans)-PIV). Multiple logistic regression analyses and linear regression modeling were conducted to determine independent predictors of the 90-day modified Rankin score (mRS) and FIV, respectively. We found that patients with favorable outcomes were younger and had lower National Institutes of Health Stroke Scale (NIHSS) score, smaller PCT-PIV, K (trans)-PIV, and smaller FIV (P < 0.001). K (trans)-PIV showed good correlation with FIV (P < 00.001, R (2) = 0.6997). In the regression analyses, K (trans)-PIV was the best predictor of clinical outcomes (P = 0.009, odds ratio (OR) = 1.960) and also the best predictor for FIV (F = 75.590, P < 0.0001). In conclusion, combining PCT and K (trans) maps derived from first-pass PCT can identify at-risk cerebral ischemic tissue more precisely than perfusion parameters alone. This provides improved accuracy in predicting FIV and clinical outcomes.
View details for DOI 10.1007/s12035-016-9838-x
View details for Web of Science ID 000399303300016
Mismatch of Low Perfusion and High Permeability Predicts Hemorrhagic Transformation Region in Acute Ischemic Stroke Patients Treated with Intra-arterial Thrombolysis.
2016; 6: 27950-?
This study sought to determine whether the permeability related parameter K(trans), derived from computed tomography perfusion (CTP) imaging, can predict hemorrhagic transformation (HT) in patients with acute ischemic stroke who receive intra-arterial thrombolysis. Data from patients meeting the criterion were examined. CTP was performed and K(trans) maps were used to assess the permeability values in HT and non-HT regions. A receiver operating characteristic (ROC) curve was calculated, showing the sensitivity and specificity of K(trans) for predicting HT risk. Composite images were produced to illustrate the spatial correlations among perfusion, permeability changes and HT. This study examined 41 patients. Twenty-six patients had hemorrhagic infarction and 15 had parenchymal hemorrhage. The mean K(trans) value in HT regions was significantly lower than that in the non-HT regions (0.26 ± 0.21/min vs. 0.78 ± 0.64/min; P < 0.001). The ROC curve analysis identified an optimal cutoff value of 0.334/min for K(trans) to predict HT risk. Composite images suggested ischemic regions with low permeability, or the mismatch area of low perfusion and high permeability, more likely have HT. HT regions after intra-arterial thrombolysis had lower permeability values on K(trans) maps. The mismatch area of lower perfusion and higher permeability are more likely to develop HT.
View details for DOI 10.1038/srep27950
View details for PubMedID 27302077
View details for PubMedCentralID PMC4908417
Delay-sensitive and delay-insensitive deconvolution perfusion-CT: similar ischemic core and penumbra volumes if appropriate threshold selected for each
2015; 57 (6): 573-581
Perfusion-CT (PCT) processing involves deconvolution, a mathematical operation that computes the perfusion parameters from the PCT time density curves and an arterial curve. Delay-sensitive deconvolution does not correct for arrival delay of contrast, whereas delay-insensitive deconvolution does. The goal of this study was to compare delay-sensitive and delay-insensitive deconvolution PCT in terms of delineation of the ischemic core and penumbra.We retrospectively identified 100 patients with acute ischemic stroke who underwent admission PCT and CT angiography (CTA), a follow-up vascular study to determine recanalization status, and a follow-up noncontrast head CT (NCT) or MRI to calculate final infarct volume. PCT datasets were processed twice, once using delay-sensitive deconvolution and once using delay-insensitive deconvolution. Regions of interest (ROIs) were drawn, and cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT) in these ROIs were recorded and compared. Volume and geographic distribution of ischemic core and penumbra using both deconvolution methods were also recorded and compared.MTT and CBF values are affected by the deconvolution method used (p < 0.05), while CBV values remain unchanged. Optimal thresholds to delineate ischemic core and penumbra are different for delay-sensitive (145 % MTT, CBV 2 ml × 100 g(-1) × min(-1)) and delay-insensitive deconvolution (135 % MTT, CBV 2 ml × 100 g(-1) × min(-1) for delay-insensitive deconvolution). When applying these different thresholds, however, the predicted ischemic core (p = 0.366) and penumbra (p = 0.405) were similar with both methods.Both delay-sensitive and delay-insensitive deconvolution methods are appropriate for PCT processing in acute ischemic stroke patients. The predicted ischemic core and penumbra are similar with both methods when using different sets of thresholds, specific for each deconvolution method.
View details for DOI 10.1007/s00234-015-1507-7
View details for Web of Science ID 000354806900004
View details for PubMedID 25749851
Using standard first-pass perfusion computed tomographic data to evaluate collateral flow in acute ischemic stroke.
Stroke; a journal of cerebral circulation
2015; 46 (4): 961-967
The study aims to determine whether volume transfer constant (K(trans)) maps calculated from first-pass perfusion computed tomographic data are a biomarker of cerebral collateral circulation and predict the clinical outcome in acute ischemic stroke caused by proximal arterial occlusion.Consecutive patients with acute occlusion of the middle cerebral artery who received endovascular treatment were enrolled. Digital subtraction angiography, computed tomographic angiography with maximum intensity projection, and K(trans) maps were used to assess their collateral circulation. Agreement between different methods was evaluated using the χ(2) tests. The correlations of various radiological and clinical outcomes with the collateral flow score, as determined from K(trans) maps, were calculated.Seventy-five patients were included, comprising 39 women and 36 men, with a mean age of 65.3±14.6 years. Collateral flow score on K(trans) maps had the highest correlation with digital subtraction angiography (κ=0.8101; P=0.9796). Twenty-five patients had poor collateral circulation on K(trans) maps, 25 had intermediate collateral flow, 20 had good collateral flow, and 5 had excellent collateral flow. Better collateral circulation was associated with better clinical outcome (P<0.0001).K(trans) maps extracted from standard first-pass perfusion computed tomography are correlated with collateral circulation status after acute proximal arterial occlusion and predictive of outcome.
View details for DOI 10.1161/STROKEAHA.114.008015
View details for PubMedID 25669309
- Using Standard First-Pass Perfusion Computed Tomographic Data to Evaluate Collateral Flow in Acute Ischemic Stroke STROKE 2015; 46 (4): 961-?
Permeability Imaging as a Biomarker of Leptomeningeal Collateral Flow in Patients with Intracranial Arterial Stenosis
CELL BIOCHEMISTRY AND BIOPHYSICS
2015; 71 (3): 1273-1279
Different methods of angiography are of great clinical utility; however, it still remains unstandardized as which method would be suitable to determine cerebral collateral circulation. Here we compared digital subtraction angiography (DSA), computer tomography angiography (CTA) and dynamic contrast-enhanced T1-weighted imaging magnetic resonance imaging (MRI) findings in seven patients with severe intracranial arterial stenosis, and determine whether volume transfer constant (K(trans)) maps of permeability imaging could be used as the biomarkers of cerebral collateral circulation. We retrospectively reviewed seven adult patients with severe intracranial arterial stenosis or occlusion with a complete parenchymal and vascular imaging work-up. DSA, CTA source imaging (CTA-SI), arterial spin labeling (ASL), and K(trans) maps were used to assess their collateral flow. Cohen's Kappa coefficient was calculated to test the consistency of their collateral scores. A reasonable agreement was found between DSA and K(trans) maps (Kappa = 0.502, P < 0.001) when all 15 regional vascular sites were included, and a better agreement found after exclusion of perforating artery territories (N = 10 sites, Kappa = 0.766, P < 0.001). The agreement between CTA-SI and DSA was moderate on all 15 sites (Kappa = 0.413, P < 0.001) and 10 sites (Kappa = 0.329, P < 0.001). The agreement between ASL and DSA was least favorable, no matter for all 15 sites (Kappa = 0.270, P < 0.001) or 10 sites (Kappa = 0.205, P = 0.002). K(trans) maps are useful and promising for leptomeningeal collateral assessment, when compared to CTA-SI or ASL. Further studies are requited for verify its validity in a large registry of patients.
View details for DOI 10.1007/s12013-014-0343-4
View details for Web of Science ID 000355342900002
Arteriopathy Diagnosis in Childhood Arterial Ischemic Stroke Results of the Vascular Effects of Infection in Pediatric Stroke Study
2014; 45 (12): 3597-?
Although arteriopathies are the most common cause of childhood arterial ischemic stroke, and the strongest predictor of recurrent stroke, they are difficult to diagnose. We studied the role of clinical data and follow-up imaging in diagnosing cerebral and cervical arteriopathy in children with arterial ischemic stroke.Vascular effects of infection in pediatric stroke, an international prospective study, enrolled 355 cases of arterial ischemic stroke (age, 29 days to 18 years) at 39 centers. A neuroradiologist and stroke neurologist independently reviewed vascular imaging of the brain (mandatory for inclusion) and neck to establish a diagnosis of arteriopathy (definite, possible, or absent) in 3 steps: (1) baseline imaging alone; (2) plus clinical data; (3) plus follow-up imaging. A 4-person committee, including a second neuroradiologist and stroke neurologist, adjudicated disagreements. Using the final diagnosis as the gold standard, we calculated the sensitivity and specificity of each step.Cases were aged median 7.6 years (interquartile range, 2.8-14 years); 56% boys. The majority (52%) was previously healthy; 41% had follow-up vascular imaging. Only 56 (16%) required adjudication. The gold standard diagnosis was definite arteriopathy in 127 (36%), possible in 34 (9.6%), and absent in 194 (55%). Sensitivity was 79% at step 1, 90% at step 2, and 94% at step 3; specificity was high throughout (99%, 100%, and 100%), as was agreement between reviewers (κ=0.77, 0.81, and 0.78).Clinical data and follow-up imaging help, yet uncertainty in the diagnosis of childhood arteriopathy remains. This presents a challenge to better understanding the mechanisms underlying these arteriopathies and designing strategies for prevention of childhood arterial ischemic stroke.
View details for DOI 10.1161/STROKEAHA.114.007404
View details for PubMedID 25388419