Increased Risk of Recurrent Stroke in Symptomatic Large Vessel Disease With Impaired BOLD Cerebrovascular Reactivity.

BACKGROUND: Impaired cerebrovascular reactivity (CVR) has been correlated with recurrent ischemic stroke. However, for clinical purposes, most CVR techniques are rather complex, time-consuming, and lack validation for quantitative measurements. The recent adaptation of a standardized hypercapnic stimulus in combination with a blood-oxygenation-level-dependent (BOLD) magnetic resonance imaging signal as a surrogate for cerebral blood flow offers a potential universally comparable CVR assessment. We investigated the association between impaired BOLD-CVR and risk for recurrent ischemic events. METHODS: We conducted a retrospective analysis of patients with symptomatic cerebrovascular large vessel disease who had undergone a prospective hypercapnic-challenged BOLD-CVR protocol at a single tertiary stroke referral center between June 2014 and April 2020. These patients were followed up for recurrent acute ischemic events for up to 3 years. BOLD-CVR (%BOLD signal change per mm Hg CO2) was calculated on a voxel-by-voxel basis. Impaired BOLD-CVR of the affected (ipsilateral to the vascular pathology) hemisphere was defined as an average BOLD-CVR, falling 2 SD below the mean BOLD-CVR of the right hemisphere in a healthy age-matched reference cohort (n=20). Using a multivariate Cox proportional hazards model, the association between impaired BOLD-CVR and ischemic stroke recurrence was assessed and Kaplan-Meier survival curves to visualize the acute ischemic stroke event rate. RESULTS: Of 130 eligible patients, 28 experienced recurrent strokes (median, 85 days, interquartile range, 5-166 days). Risk factors associated with an increased recurrent stroke rate included impaired BOLD-CVR, a history of atrial fibrillation, and heart insufficiency. After adjusting for sex, age group, and atrial fibrillation, impaired BOLD-CVR exhibited a hazard ratio of 10.73 (95% CI, 4.14-27.81; P<0.001) for recurrent ischemic stroke. CONCLUSIONS: Among patients with symptomatic cerebrovascular large vessel disease, those exhibiting impaired BOLD-CVR in the affected hemisphere had a 10.7-fold higher risk of recurrent ischemic stroke events compared with individuals with nonimpaired BOLD-CVR.

Blood oxygenation-level dependent cerebrovascular reactivity imaging as strategy to monitor CSF-hemoglobin toxicity.

OBJECTIVES: Cell-free hemoglobin in the cerebrospinal fluid (CSF-Hb) may be one of the main drivers of secondary brain injury after aneurysmal subarachnoid hemorrhage (aSAH). Haptoglobin scavenging of CSF-Hb has been shown to mitigate cerebrovascular disruption. Using digital subtraction angiography (DSA) and blood oxygenation-level dependent cerebrovascular reactivity imaging (BOLD-CVR) the aim was to assess the acute toxic effect of CSF-Hb on cerebral blood flow and autoregulation, as well as to test the protective effects of haptoglobin. METHODS: DSA imaging was performed in eight anesthetized and ventilated sheep (mean weight: 80.4 kg) at baseline, 15, 30, 45 and 60 minutes after infusion of hemoglobin (Hb) or co-infusion with haptoglobin (Hb:Haptoglobin) into the left lateral ventricle. Additionally, 10 ventilated sheep (mean weight: 79.8 kg) underwent BOLD-CVR imaging to assess the cerebrovascular reserve capacity. RESULTS: DSA imaging did not show a difference in mean transit time or cerebral blood flow. Whole-brain BOLD-CVR compared to baseline decreased more in the Hb group after 15 minutes (Hb vs Hb:Haptoglobin: -0.03 ± 0.01 vs -0.01 ± 0.02) and remained diminished compared to Hb:Haptoglobin group after 30 minutes (Hb vs Hb:Haptoglobin: -0.03 ± 0.01 vs 0.0 ± 0.01), 45 minutes (Hb vs Hb:Haptoglobin: -0.03 ± 0.01 vs 0.01 ± 0.02) and 60 minutes (Hb vs Hb:Haptoglobin: -0.03 ± 0.02 vs 0.01 ± 0.01). CONCLUSION: It is demonstrated that CSF-Hb toxicity leads to rapid cerebrovascular reactivity impairment, which is blunted by haptoglobin co-infusion. BOLD-CVR may therefore be further evaluated as a monitoring strategy for CSF-Hb toxicity after aSAH.

Foundations of Feature Selection in Clinical Prediction Modeling.

Selecting a set of features to include in a clinical prediction model is not always a simple task. The goals of creating parsimonious models with low complexity while, at the same time, upholding predictive performance by explaining a large proportion of the variance within the dependent variable must be balanced. With this aim, one must consider the clinical setting and what data are readily available to clinicians at specific timepoints, as well as more obvious aspects such as the availability of computational power and size of the training dataset. This chapter elucidates the importance and pitfalls in feature selection, focusing on applications in clinical prediction modeling. We demonstrate simple methods such as correlation-, significance-, and variable importance-based filtering, as well as intrinsic feature selection methods such as Lasso and tree- or rule-based methods. Finally, we focus on two algorithmic wrapper methods for feature selection that are commonly used in machine learning: Recursive Feature Elimination (RFE), which can be applied regardless of data and model type, as well as Purposeful Variable Selection as described by Hosmer and Lemeshow, specifically for generalized linear models.

Machine Learning Algorithms in Neuroimaging: An Overview.

Machine learning (ML) and artificial intelligence (AI) applications in the field of neuroimaging have been on the rise in recent years, and their clinical adoption is increasing worldwide. Deep learning (DL) is a field of ML that can be defined as a set of algorithms enabling a computer to be fed with raw data and progressively discover-through multiple layers of representation-more complex and abstract patterns in large data sets. The combination of ML and radiomics, namely the extraction of features from medical images, has proven valuable, too: Radiomic information can be used for enhanced image characterization and prognosis or outcome prediction. This chapter summarizes the basic concepts underlying ML application for neuroimaging and discusses technical aspects of the most promising algorithms, with a specific focus on Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), in order to provide the readership with the fundamental theoretical tools to better understand ML in neuroimaging. Applications are highlighted from a practical standpoint in the last section of the chapter, including: image reconstruction and restoration, image synthesis and super-resolution, registration, segmentation, classification, and outcome prediction.

Increased Ipsilateral Posterior Cerebral Artery P2-Segment Flow Velocity Predicts Hemodynamic Impairment.

BACKGROUND AND PURPOSE: Increased Transcranial Doppler flow velocity in the ipsilateral P2-segment of the posterior cerebral artery (PCA-P2: cm/second) is associated with recurrent cerebrovascular events in patients with unilateral internal carotid artery occlusion. However, its predictive value and correlation with hemodynamic impairment in an overall stroke patient cohort remains to be determined. METHODS: Transcranial doppler PCA-P2 flow velocity was measured in 88 patients with symptomatic unilateral steno-occlusive disease who also underwent blood oxygenation-level dependent cerebrovascular reactivity imaging (blood oxygenation-level dependent [BOLD]-cerebrovascular reactivity [CVR]). A multivariate linear regression was used to evaluate the independent correlation between the ipsilateral PCA-P2 flow velocity measurements and hemispheric BOLD-CVR. Follow-up BOLD-CVR imaging data, available in 25 patients, were used to evaluate the temporal evolution of the BOLD-CVR and PCA-P2 flow velocity association using a mixed-effect model. Furthermore, a transcranial doppler cutoff for hemodynamic failure stage 2 was determined. RESULTS: The ipsilateral systolic PCA-P2 flow velocity strongly correlated with hemispheric BOLD-CVR (R=0.79; R2=0.61), which remained unchanged when evaluating the follow-up data. Using a PCA-P2 systolic flow velocity cutoff value of 85 cm/second, patients with BOLD-CVR based hemodynamic failure stage 2 were diagnosed with an area under the curve of 95. CONCLUSIONS: In patients with symptomatic unilateral steno-occlusive disease, increased ipsilateral transcranial doppler PCA-P2 systolic flow velocity independently correlates with BOLD-CVR based hemodynamic failure. A cutoff value of 85 cm/second appears to indicate hemodynamic failure stage 2, but this finding needs to be validated in an independent patient cohort.

Impact of additional resection on new ischemic lesions and their clinical relevance after intraoperative 3 Tesla MRI in neuro-oncological surgery.

Intraoperative MRI (ioMRI) has become a frequently used tool to improve maximum safe resection in brain tumor surgery. The usability of intraoperatively acquired diffusion-weighted imaging sequences to predict the extent and clinical relevance of new infarcts has not yet been studied. Furthermore, the question of whether more aggressive surgery after ioMRI leads to more or larger infarcts is of crucial interest for the surgeons' operative strategy. Retrospective single-center analysis of a prospective registry of procedures from 2013 to 2019 with ioMRI was used. Infarct volumes in ioMRI/poMRI, lesion localization, mRS, and NIHSS were analyzed for each case. A total of 177 individual operations (60% male, mean age 45.5 years old) met the inclusion criteria. In 61% of the procedures, additional resection was performed after ioMRI, which resulted in a significantly higher number of new ischemic lesions postoperatively (p < .001). The development of new or enlarged ischemic areas upon additional resection could also be shown volumetrically (mean volume in ioMRI 0.39 cm3 vs. poMRI 2.97 cm3; p < .001). Despite the surgically induced new infarcts, mRS and NIHSS did not worsen significantly in cases with additional resection. Additionally, new perilesional ischemia in eloquently located tumors was not associated with an impaired neurological outcome. Additional resection after ioMRI leads to new or enlarged ischemic areas. However, these new infarcts do not necessarily result in an impaired neurological outcome, even when in eloquent brain areas.

Feasibility of machine learning based predictive modelling of postoperative hyponatremia after pituitary surgery.

PURPOSE: Hyponatremia after pituitary surgery is a frequent finding with potential severe complications and the most common cause for readmission. Several studies have found parameters associated with postoperative hyponatremia, but no reliable specific predictor was described yet. This pilot study evaluates the feasibility of machine learning (ML) algorithms to predict postoperative hyponatremia after resection of pituitary lesions. METHODS: Retrospective screening of a prospective registry of patients who underwent transsphenoidal surgery for pituitary lesions. Hyponatremia within 30 days after surgery was the primary outcome. Several pre- and intraoperative clinical, procedural and laboratory features were selected to train different ML algorithms. Trained models were compared using common performance metrics. Final model was internally validated on the testing dataset. RESULTS: From 207 patients included in the study, 44 (22%) showed a hyponatremia within 30 days postoperatively. Hyponatremic measurements peaked directly postoperatively (day 0-1) and around day 7. Bootstrapped performance metrics of different trained ML-models showed largest area under the receiver operating characteristic curve (AUROC) for the boosted generalized linear model (67.1%), followed by the Naïve Bayes classifier (64.6%). The discriminative capability of the final model was assessed by predicting on unseen dataset. Large AUROC (84.3%; 67.0-96.4), sensitivity (81.8%) and specificity (77.5%) with an overall accuracy of 78.4% (66.7-88.2) was reached. CONCLUSION: Our trained ML-model was able to learn the complex risk factor interactions and showed a high discriminative capability on unseen patient data. In conclusion, ML-methods can predict postoperative hyponatremia and thus potentially reduce morbidity and improve patient safety.

Machine learning in neurosurgery: a global survey.

BACKGROUND: Recent technological advances have led to the development and implementation of machine learning (ML) in various disciplines, including neurosurgery. Our goal was to conduct a comprehensive survey of neurosurgeons to assess the acceptance of and attitudes toward ML in neurosurgical practice and to identify factors associated with its use. METHODS: The online survey consisted of nine or ten mandatory questions and was distributed in February and March 2019 through the European Association of Neurosurgical Societies (EANS) and the Congress of Neurosurgeons (CNS). RESULTS: Out of 7280 neurosurgeons who received the survey, we received 362 responses, with a response rate of 5%, mainly in Europe and North America. In total, 103 neurosurgeons (28.5%) reported using ML in their clinical practice, and 31.1% in research. Adoption rates of ML were relatively evenly distributed, with 25.6% for North America, 30.9% for Europe, 33.3% for Latin America and the Middle East, 44.4% for Asia and Pacific and 100% for Africa with only two responses. No predictors of clinical ML use were identified, although academic settings and subspecialties neuro-oncology, functional, trauma and epilepsy predicted use of ML in research. The most common applications were for predicting outcomes and complications, as well as interpretation of imaging. CONCLUSIONS: This report provides a global overview of the neurosurgical applications of ML. A relevant proportion of the surveyed neurosurgeons reported clinical experience with ML algorithms. Future studies should aim to clarify the role and potential benefits of ML in neurosurgery and to reconcile these potential advantages with bioethical considerations.

Utility of deep neural networks in predicting gross-total resection after transsphenoidal surgery for pituitary adenoma: a pilot study.

OBJECTIVEGross-total resection (GTR) is often the primary surgical goal in transsphenoidal surgery for pituitary adenoma. Existing classifications are effective at predicting GTR but are often hampered by limited discriminatory ability in moderate cases and by poor interrater agreement. Deep learning, a subset of machine learning, has recently established itself as highly effective in forecasting medical outcomes. In this pilot study, the authors aimed to evaluate the utility of using deep learning to predict GTR after transsphenoidal surgery for pituitary adenoma.METHODSData from a prospective registry were used. The authors trained a deep neural network to predict GTR from 16 preoperatively available radiological and procedural variables. Class imbalance adjustment, cross-validation, and random dropout were applied to prevent overfitting and ensure robustness of the predictive model. The authors subsequently compared the deep learning model to a conventional logistic regression model and to the Knosp classification as a gold standard.RESULTSOverall, 140 patients who underwent endoscopic transsphenoidal surgery were included. GTR was achieved in 95 patients (68%), with a mean extent of resection of 96.8% ± 10.6%. Intraoperative high-field MRI was used in 116 (83%) procedures. The deep learning model achieved excellent area under the curve (AUC; 0.96), accuracy (91%), sensitivity (94%), and specificity (89%). This represents an improvement in comparison with the Knosp classification (AUC: 0.87, accuracy: 81%, sensitivity: 92%, specificity: 70%) and a statistically significant improvement in comparison with logistic regression (AUC: 0.86, accuracy: 82%, sensitivity: 81%, specificity: 83%) (all p < 0.001).CONCLUSIONSIn this pilot study, the authors demonstrated the utility of applying deep learning to preoperatively predict the likelihood of GTR with excellent performance. Further training and validation in a prospective multicentric cohort will enable the development of an easy-to-use interface for use in clinical practice.

Blood Oxygenation Level-Dependent Cerebrovascular Reactivity-Derived Steal Phenomenon May Indicate Tissue Reperfusion Failure After Successful Endovascular Thrombectomy.

In acute ischemic stroke due to large-vessel occlusion (LVO), the clinical outcome after endovascular thrombectomy (EVT) is influenced by the extent of autoregulatory hemodynamic impairment, which can be derived from blood oxygenation level-dependent cerebrovascular reactivity (BOLD-CVR). BOLD-CVR imaging identifies brain areas influenced by hemodynamic steal. We sought to investigate the presence of steal phenomenon and its relationship to DWI lesions and clinical deficit in the acute phase of ischemic stroke following successful vessel recanalization.From the prospective longitudinal IMPreST (Interplay of Microcirculation and Plasticity after ischemic Stroke) cohort study, patients with acute ischemic unilateral LVO stroke of the anterior circulation with successful endovascular thrombectomy (EVT; mTICI scale ≥ 2b) and subsequent BOLD-CVR examination were included for this analysis. We analyzed the spatial correlation between brain areas exhibiting BOLD-CVR-associated steal phenomenon and DWI infarct lesion as well as the relationship between steal phenomenon and NIHSS score at hospital discharge.Included patients (n = 21) exhibited steal phenomenon to different extents, whereas there was only a partial spatial overlap with the DWI lesion (median 19%; IQR, 8-59). The volume of steal phenomenon outside the DWI lesion showed a positive correlation with overall DWI lesion volume and was a significant predictor for the NIHSS score at hospital discharge.Patients with acute ischemic unilateral LVO stroke exhibited hemodynamic steal identified by BOLD-CVR after successful EVT. Steal volume was associated with DWI infarct lesion size and with poor clinical outcome at hospital discharge. BOLD-CVR may further aid in better understanding persisting hemodynamic impairment following reperfusion therapy.

Delayed cerebral atrophy after cerebellar stroke: topographical relation and clinical impact.

Remote dysconnectivity following cerebellar ischaemic stroke may have a negative impact on supratentorial brain tissue. Since the cerebellum is connected to the individual cerebral lobes via contralateral tracts, cerebellar lesion topography might determine the distribution of contralateral supratentorial brain tissue changes. We investigated (i) the occurrence of delayed cerebral atrophy after cerebellar ischaemic stroke and its relationship to infarct volume; (ii) whether cerebellar stroke topography determines supratentorial atrophy location; and (iii) how cortical atrophy after cerebellar stroke impacts clinical outcome. We performed longitudinal volumetric MRI analysis of patients with isolated cerebellar stroke from the Swiss Stroke Registry database. Stroke location and volume were determined at baseline MRI. Delayed cerebral atrophy was measured as supratentorial cortical volumetric change at follow-up, in contralateral target as compared to ipsilateral reference-areas. In patients with bilateral stroke, both hemispheres were analysed separately. We obtained maps of how cerebellar lesion topography, determines the probability of delayed atrophy per distinct cerebral lobe. Clinical performance was measured with the National Institutes of Health Stroke Scale and modified Rankin Scale. In 29 patients (age 58 ± 18; 9 females; median follow-up: 6.2 months), with 36 datasets (7 patients with bilateral cerebellar stroke), delayed cerebral atrophy occurred in 28 (78%) datasets. A multivariable generalized linear model for a Poisson distribution showed that infarct volume (milliliter) in bilateral stroke patients was positively associated with the number of atrophic target areas (Rate ratio = 1.08; P = 0.01). Lobe-specific cerebral atrophy related to distinct topographical cerebellar stroke patterns. By ordinal logistic regression (shift analysis), more atrophic areas predicted higher 3-month mRS scores in patients with low baseline scores (baseline score 3-5: Odds ratio = 1.34; P = 0.02; baseline score 0-2: OR = 0.71; P = 0.19). Our results indicate that (i) isolated cerebellar ischaemic stroke commonly results in delayed cerebral atrophy and stroke volume determines the severity of cerebral atrophy in patients with bilateral stroke; (ii) cerebellar stroke topography affects the location of delayed cerebral atrophy; and (iii) delayed cerebral atrophy negatively impacts clinical outcome.

Machine Learning Algorithm Identifies Patients at High Risk for Early Complications After Intracranial Tumor Surgery: Registry-Based Cohort Study.

INTRODUCTION: Reliable preoperative identification of patients at high risk for early postoperative complications occurring within 24 h (EPC) of intracranial tumor surgery can improve patient safety and postoperative management. Statistical analysis using machine learning algorithms may generate models that predict EPC better than conventional statistical methods. OBJECTIVE: To train such a model and to assess its predictive ability. METHODS: This cohort study included patients from an ongoing prospective patient registry at a single tertiary care center with an intracranial tumor that underwent elective neurosurgery between June 2015 and May 2017. EPC were categorized based on the Clavien-Dindo classification score. Conventional statistical methods and different machine learning algorithms were used to predict EPC using preoperatively available patient, clinical, and surgery-related variables. The performance of each model was derived from examining classification performance metrics on an out-of-sample test dataset. RESULTS: EPC occurred in 174 (26%) of 668 patients included in the analysis. Gradient boosting machine learning algorithms provided the model best predicting the probability of an EPC. The model scored an accuracy of 0.70 (confidence interval [CI] 0.59-0.79) with an area under the curve (AUC) of 0.73 and a sensitivity and specificity of 0.80 (CI 0.58-0.91) and 0.67 (CI 0.53-0.77) on the test set. The conventional statistical model showed inferior predictive power (test set: accuracy: 0.59 (CI 0.47-0.71); AUC: 0.64; sensitivity: 0.76 (CI 0.64-0.85); specificity: 0.53 (CI 0.41-0.64)). CONCLUSION: Using gradient boosting machine learning algorithms, it was possible to create a prediction model superior to conventional statistical methods. While conventional statistical methods favor patients' characteristics, we found the pathology and surgery-related (histology, anatomical localization, surgical access) variables to be better predictors of EPC.

BOLD cerebrovascular reactivity as a novel marker for crossed cerebellar diaschisis.

OBJECTIVE: To study blood oxygen level-dependent cerebrovascular reactivity (BOLD-CVR) as a surrogate imaging marker for crossed cerebellar diaschisis (CCD). METHODS: Twenty-five participants with symptomatic unilateral cerebrovascular steno-occlusive disease underwent a BOLD-CVR and an acetazolamide challenged (15O)-H2O-PET study. CCD and cerebellar asymmetry index were determined from PET and compared to BOLD-CVR quantitative values. Neurologic status at admission and outcome after 3 months were determined with NIH Stroke Scale (NIHSS) and modified Rankin Scale (mRS) scores. RESULTS: For both the BOLD-CVR and PET examination, a significant cerebellar asymmetry index was found for participants exhibiting CCD (CCD+ vs CCD-: for BOLD-CVR 13.11 ± 9.46 vs 1.52 ± 4.97, p < 0.001; and for PET 7.31 ± 2.75 vs 1.68 ± 2.98, p < 0.001). The area under the curve for BOLD-CVR was 0.89 (95% confidence interval: 0.75-1.0) with 0.91 sensitivity and 0.81 specificity to detect CCD. Participants exhibiting CCD were in poorer clinical condition at baseline (CCD+ vs CCD-: NIHSS 7 vs 1, p = 0.003; mRS 3 vs 1, p = 0.001) and after 3-month follow-up (NIHSS 2 vs 0, p = 0.02; mRS 1 vs 0, p = 0.04). Worse performance on both scores showed an agreement with a larger BOLD-CVR cerebellar asymmetry index. This was not found for PET. CONCLUSIONS: BOLD-CVR demonstrates similar sensitivity to detect CCD as compared to (15O)-H2O-PET in patients with symptomatic unilateral cerebrovascular steno-occlusive disease. Furthermore, participants exhibiting CCD had a poorer baseline neurologic performance and neurologic outcome at 3 months. CLASSIFICATION OF EVIDENCE: This study provides Class II evidence that BOLD-CVR identifies CCD in patients with symptomatic unilateral cerebrovascular steno-occlusive disease.