Hamam OM, Pradhan AM, Salim HA
… +14 more, Cho A, Lakhani DA, Xu R, Majmundar S, Vagal V, Hui F, Dmytriw AA, Guenego A, Nael K, Albers GW, Heit JJ, Faizy TD, Wintermark M, Yedavalli V
Distal medium-vessel occlusions (DMVOs) account for roughly 25%-40% of acute ischemic strokes and often evade early detection, delaying treatment, and worsening outcomes. Conventional imaging (non-contrast CT, CT angiogr...Distal medium-vessel occlusions (DMVOs) account for roughly 25%-40% of acute ischemic strokes and often evade early detection, delaying treatment, and worsening outcomes. Conventional imaging (non-contrast CT, CT angiography [CTA], MR angiography [MRA]) can miss smaller distal thrombi, and even experienced readers have limited sensitivity, which can be as low as 35%. Recent studies highlight that advanced neuroimaging (CT perfusion, multiphase CTA, magnetic resonance imaging [MRI]) and automated analysis improve DMVO identification. In particular, machine learning (ML) and deep learning algorithms have shown promise in detecting subtle occlusions on multimodal stroke imaging. This review summarizes current imaging approaches for DMVOs, surveys ML-based detection methods, and examines validation studies and clinical evidence. We discuss barriers to clinical integration, including the need for large, annotated datasets and regulatory validation. Finally, we outline future directions: improved algorithms (explainable AI, multimodal networks), prospective trials, and workflow integration in the neurovascular service. In sum, ML-driven DMVO detection holds potential to augment rapid stroke care, but further research and collaboration are needed to translate these tools into routine practice.
BACKGROUND AND PURPOSE: Automated detection of focal cortical dysplasia (FCD) requires large volumes of voxelwise-lesion-delineated MRI data, which are difficult to acquire. This study aims to generate synthetic MRI data...BACKGROUND AND PURPOSE: Automated detection of focal cortical dysplasia (FCD) requires large volumes of voxelwise-lesion-delineated MRI data, which are difficult to acquire. This study aims to generate synthetic MRI data exhibiting FCD, assess its realism, and evaluate its impact on automated FCD detection-particularly in reducing the need for manual annotations. METHODS: T1-weighted (T1w) and T2-weighted-fluid-attenuated inversion recovery (FLAIR) MRI scans from 131 FCD patients and 90 healthy controls from multiple (3) sites were retrospectively studied. Synthetic MRIs were generated by conditioning a generative network on binary FCD mask. Two neuroradiologists identified real images from a random set of 14 real and 14 synthetic scans. Three nnU-Net models were trained to detect FCD using (i) real-only (35-FCD/35-controls), (ii) real (35-FCD/35-controls) + synthetic augmentation, and (iii) expanded real data (70-FCD/70 controls). RESULTS: Experts showed limited ability to distinguish real from synthetic images, with classification accuracy of 60% for T1w and 70% for FLAIR (inter-rater agreement κ = 0.86). Augmenting automated FCD detection with synthetic data increased sensitivity by 8.14% (p = 0.12) and improved model confidence at true lesion sites (0.83 ± 0.11 to 0.89 ± 0.12; p = 0.02). The expanded real-data model further improved sensitivity to 73.8% (p < 0.001) and confidence to 0.90 ± 0.14 (p = 0.01). CONCLUSION: Conditional generative networks can generate realistic synthetic FCD-MRIs, reducing labeled data needs by ∼20% while maintaining equivalent sensitivity. Equivalent amounts of real data, when available, remain more effective than synthetic augmentation.
BACKGROUND AND PURPOSE: Stroke is a leading cause of epilepsy, especially in older adults. The SeLECT score remains the standard among post-stroke epilepsy (PSE) prediction tools. However, its broader validation is limit...BACKGROUND AND PURPOSE: Stroke is a leading cause of epilepsy, especially in older adults. The SeLECT score remains the standard among post-stroke epilepsy (PSE) prediction tools. However, its broader validation is limited by the need to manually extract neuroimaging predictors (cortical and middle cerebral artery [MCA] involvement). Unlike the CAVE score, SeLECT did not evaluate acute stroke volume, which can now be quantified automatically. We aimed to determine whether stroke volume independently predicts PSE and compare its predictive contribution to SeLECT's neuroimaging variables. METHODS: SeLECT variables were manually extracted. Diffusion-weighted imaging volume was quantified using a validated convolutional neural network. Cox proportional hazards models for time to PSE were built by adding stroke volume (per 10 mL) and then removing cortical and/or MCA involvement. For each model, we analyzed variable significance, discrimination, and calibration. RESULTS: Among 221 patients, 35 (15.8%) developed PSE. In our cohort, the original SeLECT score and the refit model had a C-index of 0.669 and 0.642, respectively. Adding stroke volume resulted in a C-index of 0.656. Retaining volume while removing cortical and MCA involvement resulted in C-indices of 0.664 and 0.668, respectively. Keeping stroke volume and removing both variables increased the C-index to 0.679. Calibration was good for all models. Stroke volume in crease by 10 mL was an independent predictor of 12% increased PSE risk across all models. CONCLUSIONS: Acute stroke volume is an independent PSE predictor. Stroke volume offered comparable discrimination to the neuroimaging components of the SeLECT score, supporting its use as a scalable and automated alternative.
BACKGROUND: Facial nerve (CN VII) diffusion MR tractography is considered as a useful adjunct in pre-operative planning prior to vestibular schwannoma (VS) resection, especially in larger (Koos Grade III/IV) tumors. Sinc...BACKGROUND: Facial nerve (CN VII) diffusion MR tractography is considered as a useful adjunct in pre-operative planning prior to vestibular schwannoma (VS) resection, especially in larger (Koos Grade III/IV) tumors. Since 2016, several systematic reviews have investigated the clinical value of CN VII tractography in VS, and all reported a "success rate" of at least 87% for predicting the pre-operative CN VII position. Yet in clinical practice, CN VII tractography has not yet been widely adopted into routine clinical practice. We suspected that underlying methodology and reporting metrics for existing tractography algorithms may be overestimating success rate. This motivated us to revisit the literature from a different perspective to unravel the caveats and nuances behind this technology. METHODS: We screened all published works on PubMed related to pre-operative CN VII tractography in VS. Twenty-two studies were reviewed in detail. RESULTS: We observed a strikingly high heterogeneity in tractography protocols in all domains of the tractography acquisition and analysis pipeline across studies. CONCLUSIONS: These findings suggest that the reliability and reproducibility of CN VII tractography in large VS has been overestimated. We believe that employing standardized reporting metrics, including sensitivity, true predictive value, and false discovery rate, would increase the transparency of benchmarking over other commonly reported metrics ("success rate" or "concordance rate"). In addition, ongoing research should aim to systematically investigate and improve each step in the acquisition and analysis pipeline for CN VII tractography in VS.
BACKGROUND: Recent studies suggest that disruptions of the blood-cerebrospinal fluid (CSF) barrier within the choroid plexus (ChP) may contribute to multiple sclerosis (MS) pathogenesis. We investigated the relationship...BACKGROUND: Recent studies suggest that disruptions of the blood-cerebrospinal fluid (CSF) barrier within the choroid plexus (ChP) may contribute to multiple sclerosis (MS) pathogenesis. We investigated the relationship between a quantitative marker of ChP enhancement and markers of focal and diffuse brain tissue injury in MS. METHODS: A group of 34 MS participants and 21 healthy participants underwent 7T MRI including magnetization prepared 2 rapid acquisition gradient echoes (MP2RAGE) and fluid attenuated inversion recovery (FLAIR) acquisitions. The MS group received contrast, and delta T1 (ΔT1) maps were computed to assess enhancement. ChP, white matter lesions (WML), normal-appearing white matter (NAWM), and gray matter (GM) were segmented. Pre-contrast quantitative T1 (qT1) values were compared between groups, and linear regression with mean ChP ΔT1 was performed for WML volume and pre-gadolinium (Gd) mean qT1 of WML, NAWM, and GM. RESULTS: Mean qT1 of ChP, NAWM, and GM, as well as ChP volume, were higher in MS compared to controls (p < 0.001). ChP ΔT1 was significantly associated with pre-Gd qT1 of NAWM (β = 0.20, R = 0.54, p < 0.001) and GM (β = 0.18, R = 0.49, p < 0.001), but not WML volume (p = 0.3) or WML qT1 (p = 0.05). CONCLUSIONS: The association between ChP enhancement and diffuse tissue injury, together with elevated qT1 values and ChP volumes in MS, supports a mechanism of brain injury involving CSF-mediated toxicity distinct from classic lesion pathology in MS.
Le NM, Samaha J, Iyyangar AS
… +13 more, Gomez-Farias J, Kfoury B, Bajaj R, Nguyen C, Shams S, Neal-Harris C, Ebirim E, Azeem H, Ballekere AN, Dhanjani S, Lee E, Giancardo L, Sheth SA
BACKGROUND AND PURPOSE: We evaluated agreement and performance of non-contrast head-computerized tomography (NCHCT) and CT-perfusion (CTP) in identifying large core infarct in acute ischemic stroke (AIS) due to large ves...BACKGROUND AND PURPOSE: We evaluated agreement and performance of non-contrast head-computerized tomography (NCHCT) and CT-perfusion (CTP) in identifying large core infarct in acute ischemic stroke (AIS) due to large vessel occlusion (LVO) undergoing endovascular therapy (EVT), using MRI as reference. METHODS: From our prospective multicenter registry, we identified patients with LVO-AIS due to internal carotid artery or middle cerebral artery M1occlusions who underwent EVT between 2017 and 2024. Final infarct volume (FIV) was defined using 24-48 h post-EVT diffusion-weighted imaging magnetic resonance imaging (MRI-FIV). To limit infarct growth bias, only patients with CTP-to-EVT start time <3 h were included. Large core infarct was defined at FIV thresholds: 50, 70, and 100 mL. The primary outcome was agreement between NCHCT and CTP in identifying large core infarct using kappa-statistics. Large core was considered if NCHCT-ASPECTS<6 or rCBF<30% volume>70 mL on CTP (RAPID/Viz.AI). Secondary outcomes included classification accuracy of each modality relative to MRI-FIV using the area under the receiver operating characteristic curve (AUC-ROC). Sensitivity analyses were performed in subgroups with TICI 2c-3 and cases processed by RAPID. RESULTS: Among 241 EVT-treated LVO-AIS patients, median NIHSS was 15 [IQR: 10-20], MRI-FIV 13.8 Ml [IQR: 5-41.0], ASPECTS 8 [IQR: 7-10], and CTP-predicted core 8 mL [IQR: 0-31.0]. CTP and NCHCT showed slight agreement in identifying large core (κ = 0.192) and weak-to-acceptable discrimination for identifying large core infarcts (AUC-ROC: 0.61-0.72 across MRI-FIV thresholds). Both modalities showed limited predictive ability for 90-day functional independence (AUC-ROC: 0.63-0.65). Similar findings were observed in sensitivity analyses. CONCLUSIONS: Among LVO-AIS EVT-treated patients, NCHCT and CTP demonstrated slight agreement in classifying small versus large core, and neither technique was effective at predicting FIV or clinical outcomes.
BACKGROUND AND PURPOSE: Resting state functional connectivity can be measured using resting state functional MRI (fMRI), but also task-dependent fMRI in blocked designs. The latter has been demonstrated in healthy partic...BACKGROUND AND PURPOSE: Resting state functional connectivity can be measured using resting state functional MRI (fMRI), but also task-dependent fMRI in blocked designs. The latter has been demonstrated in healthy participants but not yet validated in clinical cohorts. Since functional connectivity of resting state networks (e.g., default mode network [DMN] and somatomotor network [SMN]) is altered in people with epilepsy, and the impact of the disease on the quality of the intermittent resting state data is unclear, we aimed to validate the method using a clinical fMRI in people with epilepsy. METHODS: We compared functional connectivity derived from a standard resting state with rest periods of a clinical language fMRI (intermittent resting state) of 92 people with focal epilepsy. Both methods were analyzed across different aspects of functional connectivity: topography, within-network connectivity, and group-level comparisons. Therefore, we conducted independent component analyses (ICAs), similarity-, regions of interest (ROI)-to-ROI-, and second-level seed-based analyses. RESULTS: Results indicated similar ICA-derived topography of DMN and SMN from both methods. Within-network connectivity also yielded comparable results. Seed-based analyses of left and right hippocampal connectivity in people with left and right temporal lobe epilepsy also revealed analogous results, with minor restrictions in right hippocampal connectivity. CONCLUSION: The intermittent resting state method produces highly similar results to a standard resting state method in people with epilepsy across different aspects of functional connectivity. It is, therefore, an efficient approach to gain insights into functional connectivity networks in a clinical cohort without performing an additional resting state fMRI.
BACKGROUND AND PURPOSE: Recent MRI developments have allowed for in vivo myelin imaging in clinically feasible time frames. This retrospective study aimed to evaluate the ability of the Rapid Estimation of Myelin for Dia...BACKGROUND AND PURPOSE: Recent MRI developments have allowed for in vivo myelin imaging in clinically feasible time frames. This retrospective study aimed to evaluate the ability of the Rapid Estimation of Myelin for Diagnostic Imaging (REMyDI) technique in monitoring longitudinal myelin changes and brain atrophy in persons with multiple sclerosis (pwMS) undergoing treatment with rituximab or autologous hematopoietic stem cell transplantation (aHSCT). METHODS: Between May 2017 and January 2022, 62 pwMS treated with either rituximab (n = 25) or aHSCT (n = 37) underwent brain MRI scans at three time points. A 3 Tesla brain MRI was performed, including 3D T1-weighted imaging, 3D T2-weighted fluid-attenuated inversion recovery imaging, and 2D multi-dynamic multi-echo imaging for REMyDI and brain volumetrics. Longitudinal changes in imaging parameters and associations with the Expanded Disability Status Scale and Symbol Digit Modalities Test were analyzed using mixed-effects models. RESULTS: The rituximab group exhibited increases in whole-brain myelin (+0.25 mL per year), cortical myelin (+0.11 mL per year), and myelin in normal-appearing deep gray matter (NADGM) (+0.02 mL per year). In contrast, these measures were stable or declined in the aHSCT group. Brain parenchymal fraction showed a larger reduction in the rituximab group (-0.68% per year) compared to the aHSCT group (-0.24% per year). Myelin-related imaging measures showed positive but nonsignificant associations with clinical parameters. CONCLUSIONS: REMyDI enables longitudinal assessment of myelin-related metrics in vivo, which complements conventional brain volumetrics and is suitable for monitoring treatment responses in MS.
Pineda-Ibarra C, Puig J, Nuñez-Leiva D
… +11 more, Rodríguez Y, Mora J, Medrano-Martorell S, Pariente JC, Comas-Cufí M, Niñerola-Baizán A, Farré-Melero A, Reyes-Barrios DA, Aldecoa I, Oleaga L, González-Ortiz S
BACKGROUND AND PURPOSE: Grading meningioma guides treatment choices from follow-up to surgical resection with adjuvant radiation. Radiomics may offer a non-invasive alternative to biopsies. We assessed radiomic features...BACKGROUND AND PURPOSE: Grading meningioma guides treatment choices from follow-up to surgical resection with adjuvant radiation. Radiomics may offer a non-invasive alternative to biopsies. We assessed radiomic features (RFs) for distinguishing Grade 1 and Grade 2 meningiomas on preoperative multiparametric MRI. METHODS: Presurgical T1-weighted (T1), T2-weighted (T2), T2 gradient echo-weighted (T2GRE), fluid-attenuated inversion recovery (FLAIR), apparent diffusion coefficient (ADC), and T1-weighted contrast-enhanced (T1CE). MRI sequences of histopathologically diagnosed meningiomas were collected retrospectively. Each volume had 75 RFs extracted from semimanually segmented tumors using MintLesion Research (Version 3.10). The Lasso method selected variables from imputed data, and 10-fold cross-validation determined the optimal regularization parameter. For Lasso-retained variables, multivariate effects were estimated. RESULTS: Out of 150 patients (67.3% women), 110 (73.3%) had Grade 1 meningiomas, and 40 (26.7%) Grade 2. The strongest metrics to distinguish meningiomas Grade 1 versus Grade 2 were intensity histogram coefficient of variation on T1CE (odds ratio [OR] 0.47, 95% confidence interval [CI] 0.23-0.88; p = 0.028), maximum histogram gradient on T1 (OR 2.11, 95% CI 1.18-4.82; p = 0.043), and intensity histogram quartile coefficient of dispersion on FLAIR (OR 0.53, 95% CI 0.31-0.89; p = 0.021). The combined RFs achieved an area under the curve of 0.814 (95% CI, 0.732-0.896) for grading differentiation. Texture features and metrics extracted from T2, T2GRE, and ADC sequences did not discriminate meningioma grading. CONCLUSIONS: Histogram-based first-order RFs from T1, FLAIR, and T1CE may predict meningioma grades preoperatively. Larger, multicenter studies are needed to confirm these findings, providing insights for clinical decision-making and personalized treatment.
BACKGROUND: Previous studies have identified hemispheric asymmetries in cerebral blood flow and volume, favoring the left hemisphere. Accordingly, we hypothesized that arteries on the left side of the circle of Willis (C...BACKGROUND: Previous studies have identified hemispheric asymmetries in cerebral blood flow and volume, favoring the left hemisphere. Accordingly, we hypothesized that arteries on the left side of the circle of Willis (CoW) are larger than on the right. We compared artery diameters between the hemispheres. METHODS: Cranial time-of-flight magnetic resonance angiography scans of 1052 participants from a population-based cohort were assessed. Diameters of major CoW arteries (> 1.2 mm) were measured using a semiautomatic tool (mean ± standard deviation) and compared between the left and right hemisphere using a paired-samples t-test. As the posterior communicating arteries (Pcom) are often small and non-normally distributed, they were measured manually, categorized as "present" (≥ 1 mm) or "aplastic/hypoplastic" (< 1 mm), and compared using odds ratios (OR) with 95% confidence intervals (CI). RESULTS: The A2 segment of the anterior cerebral artery was smaller on the left than on the right (1.97 ± 0.21 mm vs. 2.01 ± 0.22 mm; p < 0.001), while the vertebral artery (2.36 ± 0.44 mm vs. 2.25 ± 0.41 mm; p < 0.001) and P1 segment of the posterior cerebral artery (2.01 ± 0.28 mm vs. 1.98 ± 0.29 mm; p = 0.001) were larger on the left. The Pcom was less frequently present on the left (26.7%) than on the right (33.4%; OR 0.73, 95% CI 0.60-0.88). No left-right differences were found for the A1 segment, M1 segment of the middle cerebral artery, and internal carotid artery. CONCLUSIONS: We found that some vessels were larger in the left hemisphere, whereas others were smaller. Future studies should investigate underlying mechanisms driving these specific asymmetries.
BACKGROUND AND PURPOSE: The amygdala plays a key role in the pathophysiology of autoimmune limbic encephalitis (ALE), contributing to epileptic seizures and neuropsychiatric symptoms. While no study has examined microstr...BACKGROUND AND PURPOSE: The amygdala plays a key role in the pathophysiology of autoimmune limbic encephalitis (ALE), contributing to epileptic seizures and neuropsychiatric symptoms. While no study has examined microstructural changes in individual amygdala nuclei in ALE, we used the T1-weighted/T2-weighted (T1w/T2w) ratio to explore amygdalar pathology and its associations with clinical manifestations, including epilepsy and neuropsychiatric symptoms. METHODS: This single-center study examined 57 patients diagnosed with ALE and 16 healthy controls (HC). Patients underwent a comprehensive assessment that included clinical, electroencephalogram (EEG), magnetic resonance imaging (MRI), and neuropsychological assessments. Patients were stratified by epileptic focus based on long-term EEG. T1w/T2w ratio and volumetric measures of the amygdala and its nuclei were analyzed and correlated with epileptic focus and neuropsychiatric outcomes. RESULTS: EEG revealed 26 left temporal, 26 bitemporal, and five right temporal epileptic foci. The T1w/T2w ratio in the left amygdala was markedly reduced in patients with left temporal (p = 0.013) and bitemporal (p = 0.018) epileptic foci compared to HC. This reduction was most pronounced in the left basolateral complex (p = 0.011). Whereas amygdalar volumes were similar between patients and HC, exploratory analyses showed an increased volume of the left lateral nucleus in left temporal ALE (p = 0.036). Furthermore, we found no correlations between MRI measures and neuropsychiatric scores. CONCLUSION: Our findings indicate that the basolateral complex of the amygdala is preferentially affected in ALE, suggesting a region-specific vulnerability to autoimmune-mediated inflammation. T1w/T2w ratio alterations reflect the epileptogenic focus and may serve as a clinically accessible, noninvasive biomarker for early diagnosis and treatment monitoring in ALE.
BACKGROUND AND PURPOSE: Embolic stroke of undetermined source (ESUS) may be associated with carotid artery plaques with <50% stenosis. Plaque vulnerability is multifactorial, possibly related to intraplaque hemorrhage (I...BACKGROUND AND PURPOSE: Embolic stroke of undetermined source (ESUS) may be associated with carotid artery plaques with <50% stenosis. Plaque vulnerability is multifactorial, possibly related to intraplaque hemorrhage (IPH), lipid-rich necrotic core, perivascular adipose tissue (PVAT), and calcifications. Machine learning (ML)-based plaque classification is increasingly popular but often limited in clinical interpretability by black-box nature. We applied an explainable ML approach, using noncalcified plaque components and calcification features with the SHapley Additive exPlanations (SHAP) framework to classify plaques as culprit or nonculprit. METHODS: This was a retrospective, cross-sectional study. Patients with unilateral anterior circulation ESUS with calcified carotid plaques in neck computed tomography (CT) angiography were analyzed. Calcification-level features were derived from manual segmentations. Plaque-level features were assessed by a neuroradiologist and by semi-automated software. Plaques were classified as culprit if ipsilateral to stroke side. Eight classifiers were benchmarked, and a gradient-boosted decision tree (CatBoost) was further tuned. SHAP explained model decisions. RESULTS: Seventy patients yielded 116 calcified plaques (270 calcifications). Model based on five plaque- and calcification-level features achieved ROC-AUC (receiver operating characteristic area under the curve) 0.79 and precision-recall-AUC 0.86, outperforming classification based on plaque thickness ≥3 mm (ROC-AUC 0.59, p = 0.04) and IPH presence (ROC-AUC 0.51, p = 0.003). SHAP identified plaque thickness and PVAT volume as the most influential features with potential thresholds of >2.6 mm and ≥112 mm, respectively.f CONCLUSIONS: ML model trained with noncalcified plaque and calcification features can classify culprit calcified carotid plaque better than conventional criteria. Using clinically interpretable features with SHAP, the model explained its decisions and suggested hypothesis-generating thresholds.
Lapucci C, Gambella M, Cipriano E
… +10 more, Raiola AM, Varaldo R, Ghiso A, Centanaro M, Capello E, Schenone A, Castellan L, Barletta L, Angelucci E, Inglese M
BACKGROUND AND PURPOSE: Chimeric antigen receptor-engineered T-cell (CAR-T) therapy in hematological malignancies may be associated with severe complications, as Cytokine Release Syndrome (CRS) and Immune effector Cell-A...BACKGROUND AND PURPOSE: Chimeric antigen receptor-engineered T-cell (CAR-T) therapy in hematological malignancies may be associated with severe complications, as Cytokine Release Syndrome (CRS) and Immune effector Cell-Associated Neurotoxicity Syndrome (ICANS). The aim of the study is to investigate MRI-derived macrostructural and microstructural features potentially able to identify patients at higher ICANS risk. METHODS: Forty-two patients treated with CAR-T from October 2020 to June 2025 performed brain MRIs before CAR-T administration, including diffusion-weighted imaging. A general linear model was used to compare patients who developed ICANS, CRS, or neither at baseline in terms of MRI macro- and microstructural features. A binary logistic regression analysis was performed to evaluate the role of microstructural features in predicting the risk of developing ICANS. RESULTS: Mean age 59.2 ± 13 years, 59.5% male; 21 (50%) patients received tisagenlecleucel, 21 (50%), axicabtagene ciloleucel or brexucabtagene autoleucel; 14 (33%) and 31 (73.8%) patients developed ICANS and CRS, respectively. At baseline MRI, fluid-attenuated inversion recovery (FLAIR) white matter (WM) hyperintensities were detected in 41/42 (97.6%). No significant differences between patients who developed ICANS, CRS and neither both were observed in terms of FLAIR hyperintensities nor total brain volume at baseline. Fractional anisotropy extracted from FLAIR hyperintensities and WM areas without macroscopic abnormalities was a predictor of ICANS in the logistic regression model (p = 0.03 and 0.02, respectively). CONCLUSIONS: FLAIR hyperintensities and brain volume prior to CAR-T were not informative, whereas the severity of WM microstructural (axonal) damage predicted ICANS risk. Greater axonal damage was associated with a higher likelihood of ICANS.