Large language models (LLMs) are increasingly used in medicine and research, but neuroradiologists’ awareness, perceived utility, and concerns about integrity and disclosure remain incompletely characterized. This survey aimed to assess radiologists’ awareness and perceptions of LLMs in clinical and research domains. An anonymous, voluntary SurveyMonkey survey was distributed internationally (October 1, 2024 to March 31, 2025) via neuroradiology society newsletters/membership channels and social media. Categorical variables were summarized as counts and percentages; Likert items were summarized using weighted means and response distributions. Item-level complete-case denominators were reported. Prespecified subgroup analyses used chi-square/Fisher exact tests (categorical) and nonparametric tests (ordinal), with Holm multiplicity control within prespecified multi-item question blocks and within each subgroup factor. A total of 265 respondents started the survey; after exclusions, 209 were included in the analytic sample, of whom 64.6
Pure arterial malformations (PAMs) are rare arterial anomalies typically considered incidental, and their role in acute large vessel occlusion is not well defined. We describe a case of acute large vessel occlusion occurring in direct anatomical association with a carotid terminus PAM, treated with mechanical thrombectomy and followed by early selective re-occlusion of the malformed segment. Multimodal imaging and clinical history suggested pre-existing hemodynamic compromise and in situ thrombosis with distal embolization. Mechanical thrombectomy across the PAM segment was technically feasible and resulted in distal reperfusion, but durable patency of the malformed tract was not maintained. This case expands the recognized ischemic spectrum of PAM and suggests that selected morphologically complex lesions may behave as hemodynamically unstable substrates rather than incidental vascular variants.
Through its global and extensive connections with multiple brain structures, the thalamus can play an important role in seizure propagation. This study investigated volumetric and functional connectivity alterations of the thalamus and more specifically the thalamic nuclei subgroups using 7T MRI in patients with MRI-negative drug-resistant focal epilepsy (DRE). Twenty-nine patients with MRI-negative DRE and 49 healthy controls underwent 7T MRI, including T1-weighted and resting-state functional MRI sequences. Thalamic nuclei segmentations were derived using automated segmentation tools and split into anterior (ANT), medial, ventral, and posterior nuclei groups. Functional connectivity with the rest of the brain was assessed using degree centrality (DC). Linear regression models evaluated group differences and relationships with epilepsy characteristics while controlling for age and sex. Asymmetry indices assessed the ipsi- vs. contralateral differences in the patient cohort. Patients showed decreased bilateral medial and increased bilateral ventral relative volumes, alongside a right-sided ANT volume increase, compared with controls. DC was increased in the left posterior and ventral nuclei. Both volume and DC in the medial, ventral, and whole-thalamic ROIs were related to epilepsy duration, while ventral and whole-thalamic DC asymmetry distinguished patients by seizure type and seizure duration. Both structural and functional alterations of the thalamic nuclei groups in patients with DRE were observed. Ventral, medial, and posterior nuclei metrics show promise as imaging biomarkers of disease burden and seizure severity, thus encouraging further research into both structural and functional properties of thalamic nuclei in focal epilepsy.
Growing evidence has linked MRI-visible perivascular spaces (PVS) to multiple sclerosis (MS) pathogenesis. However, whether PVS are associated with increased microstructural damage in white matter lesions (WML) remains unclear. Using diffusion kurtosis imaging (DKI), we aimed to explore the associations between PVS and the microstructural damage of WML in relapsing-remitting multiple sclerosis (RRMS) people with or without disease-modifying therapies (DMT) and their correlations with clinical biomarkers of disability and cognitive function. The study included 68 people with RRMS and 47 age- and sex-matched HCs. WML were categorized into four groups based on two factors: whether the lesions were penetrated by perivascular spaces (P_WML vs. NP_WML) through visual assessment, and whether patients received DMT (MS + vs. MS−). The diffusion metrics, including fractional anisotropy (FA), kurtosis fractional anisotropy (KFA), mean diffusivity (MD) and mean kurtosis (MK), were normalized using ROI-specific z-scores derived from HCs and used to assess the microstructural damage in WML. Lesion-wise group comparisons were performed using linear mixed-effects models and Spearman partial correlation analysis between diffusion metrics of WML and cognitive performance and clinical disability were performed. In the MS- and MS+ groups, the MKindex in P_WML was significantly lower than NP_WML (MS-: estimate [95
Presurgical functional mapping is essential for maximizing tumor resection while preserving motor and language function. While task-based fMRI (tb-fMRI) is the current non-invasive standard, its utility is substantially limited in patients with aphasia, paresis, or cognitive impairment who cannot comply with active paradigms. Resting-state fMRI (rs-fMRI) offers a task-free alternative, yet its spatial fidelity and clinical validity compared to task-based methods remain under rigorous evaluation. In this prospective study, we evaluated 12 patients with histologically confirmed brain tumors who underwent rs-fMRI and tb-fMRI for preoperative sensorimotor and language mapping. Motor mapping utilized a bilateral finger tapping block-design paradigm. Language mapping utilized three distinct task paradigms (Silent Word Generation, Rhyming, and Synonym). Spatial concordance was quantified using Dice Similarity Coefficients (DSC), Jaccard Indices, and Receiver Operating Characteristic (ROC) analysis. rs-fMRI demonstrated fair to moderate spatial concordance with tb-fMRI. For motor mapping, the mean DSC was 0.62 ± 0.08 with high specificity (0.93 ± 0.02). In the language domain, rs-fMRI showed the highest concordance with the Silent Word Generation task (DSC 0.63 ± 0.11). Notably, rs-fMRI exhibited superior signal stability, with significantly fewer motion-censored volumes compared to language task paradigms (p < 0.01). While rs-fMRI demonstrated moderate sensitivity ( 0.62), it achieved high specificity (0.93–0.94) across all domains however the moderate sensitivity indicates that functionally important regions may remain undetected. In this preliminary single-center study, rs-fMRI demonstrates fair to moderate spatial concordance with tb-fMRI and high specificity in delineating core eloquent cortex, though with moderate sensitivity that may leave functionally important regions undetected. The study provides hypothesis-generating evidence supporting the integration of rs-fMRI into multimodal presurgical workflows for neuro-oncology patients who cannot comply with task-based paradigms.
V1 vertebral artery (VA) tortuosity is associated with cervical artery dissection, stent fracture, and endovascular access difficulty, and predominates on the left; its anatomical basis and its relationship to VA dominance have not been quantitatively tested on CT angiography (CTA). Bilateral V1 morphometric analysis was performed on 50 cervical CTAs (24 men, 26 women; 63.2 ± 15.5 years). The tortuosity index (TI) was computed as the centreline-to-straight-line length ratio; VA dominance was determined at two segmental levels. Left V1 was more tortuous than right (TI 1.37 ± 0.26 vs. 1.20 ± 0.15; Wilcoxon p < 0.0001; d_z = 0.74) in 82
The relative contribution of venous outflow obstruction and intracranial pressure (ICP) to symptomatology and treatment response in Idiopathic Intracranial Hypertension (IIH) remains uncertain. Venous sinus stenting (VSS) is indicated in patients with significant trans-stenotic pressure gradients (TPG) and concordant symptoms. A subset of these patients demonstrates conventionally normal superior sagittal sinus (SSS) pressures. As SSS pressure is often interpreted as a surrogate for ICP, it remains uncertain whether these patients derive benefit from VSS. A multicentre VSS registry was retrospectively queried to identify IIH patients with normal SSS pressure (≤ 15 mmHg). These patients were matched 1:1 to patients with elevated SSS pressure (> 15 mmHg) based on pre-stent TPG. Primary outcomes were improvement in headache, tinnitus and papilledema. Multivariable regression assessed the association between SSS pressure and clinical outcomes, adjusting for residual differences in pre-stent TPG. Among 517 patients undergoing VSS, 33 (6.4
Randomized evidence for mechanical thrombectomy (MT) in medium vessel occlusions, particularly beyond 6 h, is heterogeneous. In a secondary analysis of a previously reported single-center cohort, we compared MT for isolated M2/M3 occlusions within versus beyond 6 h from last known well. Consecutive patients treated between January 2020 and August 2023 were retrospectively stratified by last-known-well-to-groin-puncture time (≤ 6 h vs. > 6 h). The primary outcome was functional independence (modified Rankin Scale [mRS] 0–2) at 90 days. Because all patients underwent MT, superiority over medical management cannot be addressed. Logistic regression was adjusted for age and baseline NIHSS. Seventy-six patients were analyzed (56 early, 20 late). No significant difference was detected in 90-day functional independence (8/20 [40.0
Mechanical thrombectomy (MT) is the first-line therapy for patients with acute ischemic stroke due to large vessel occlusion; however, despite substantial advances in MT techniques and devices, mechanical recanalization remains unsuccessful in up to 15
Standard assessment of treatment response in craniopharyngioma (CP) includes volumetric measurement of solid and cystic tumour components on contrast-enhanced MRI. This study compared the detection of tumour progression using MRI with and without contrast administration. Progressive disease (PD) was assessed based on the measurement of tumour volume change using MRI without contrast. Contrast-enhanced MRI served as the reference standard. In most patients, non-contrast T1- and T2-weighted sequences were sufficient to assess PD. Non-contrast MRI showed a sensitivity of 89.9
Although radiological severity thresholds for amyloid-related imaging abnormalities with hemosiderin deposition (ARIA-H) were developed using 2-dimensional acquisitions of T2*-weighted gradient-recalled echo (T2*GRE) MRI in pivotal anti-amyloid trials, adoption of well established 3D susceptibility-weighted imaging (SWI) in ARIA-H monitoring in clinical practice represents a reasonable long-term direction for clinical practice. SWI provides documented higher sensitivity and comparable or superior inter-rater reliability compared with conventional T2*GRE for detecting cerebral microbleeds and cortical superficial siderosis, the two key imaging manifestations of ARIA-H. Moreover, SWI is already incorporated into most contemporary dementia MRI protocols, offering important practical and logistical advantages. Available preliminary evidence suggests that the downstream clinical impact of detecting a few additional microbleeds or areas of siderosis may remain modest for most patients. Earlier and more reliable detection could potentially enhance safety by identifying individuals at higher risk of ARIA-H before treatment initiation. The critical requirements are transparency regarding sequence choice and consistency within individual patients over time.
T1 weighted (T1w), T2 weighted (T2w) and T1w/T2w have been used for perivascular space (PVS) segmentation. However, studies on the relationships or comparisons across them are few with limited sample sizes and yield inconsistent results. Examining those relationships can help to understand the consistency and reliability of PVS measurements for each method. This study systematically examined the relationships among PVS volumes and counts derived from T1w, T2w, and combined T1w/T2w methods with 2,654 participants. PVS segmentation was performed using Frangi’s filter. Measurement repeatability and sensitivity to physiological and pathological differences were compared across the three approaches. Strong correlations were observed for T1w (R² = 0.89–0.96) and T2w (R² = 0.98–0.99) relative to the T1w/T2w method, as well as T1w vs. T2w methods (R²=0.87–0.95). All three methods showed robust repeatability with no significant differences observed in test-retest analyses. Significantly higher PVS (including volume and count) was found in males. All three methods could also identify physiological and pathological variations. Significant differences were observed between participants with poor and good sleep quality, as well as between participants with hypertension and non-hypertension across all methods. However, the T1w method demonstrated weaker group separation, with smaller mean differences in PVS measurements than those observed using the T2w and T1w/T2w methods. The findings support the use of all three methods for automated PVS quantification in clinical and research settings, while T2w and combined T1w/T2w approaches potentially detect more PVS.
Transdural supply (TDS) is a recognized angioarchitectural feature of brain arteriovenous malformations (bAVMs). We aimed to systematically characterize the angioarchitecture of TDS in bAVMs, focusing on whether transdural feeders terminated within the nidus or connected directly to the draining vein. TDS prevalence and its associated clinical features were also investigated. This retrospective study enrolled 521 patients (524 bAVMs) from 16 centers who underwent systematic six-vessel digital subtraction angiography. bAVMs were classified as nidus- or fistula-dominant. Angiography was evaluated with specific focus on the presence of TDS and the termination site of transdural feeders. TDS was classified as TDS-Nidus (fistulous point within the nidus) or TDS-DV (fistulous point at the draining vein wall). TDS was identified in 88 bAVMs (16.8
Differentiating tandem from isolated internal carotid artery (ICA) occlusion is important for guiding treatment to achieve the best outcomes, as tandem lesions are associated with more complex strategies and potentially worse outcomes. Although computed tomography angiography (CTA) is widely used, diagnostic challenges such as pseudo-occlusions—false-positive imaging findings—limit reliability. This study evaluates whether quantitative imaging parameters can support the distinction. This retrospective, single-center study included 71 patients diagnosed with ICA occlusion between 2020 and 2024. 35 patients in the tandem group and 36 in the isolated ICA occlusion group were included. The demographic information (age, sex, occlusion etiology, and comorbidity) and CTA-based imaging parameters were collected. Predictors were selected among candidates using univariate, penalized regression, and pairwise Pearson correlation analyses. The primary three-predictor model was estimated using Firth penalized logistic regression and evaluated using internal validation and calibration. Demographic data were comparable among study groups, except age. Following the multistep predictor-selection process, the mean density of the occluded ICA pre-occlusion (Mean_Occl_ICA_pre), the mean densities of the thrombus permeability ratio (Thrm_Perm_Ratio_Mean), and the gradient ratio of the occluded ICA post-occlusion (GR_Occl_ICA_post) were included in the final model. The model showed a bootstrap optimism-corrected AUC of 0.713. The combination of CTA-derived attenuation parameters showed modest discrimination between tandem and isolated ICA occlusion. These parameters may provide complementary diagnostic information, but validation in larger independent cohorts is required.
This study sought to quantify, through a multi-reader study, whether AI assistance improves diagnostic accuracy across experience levels, reduces bidirectional errors, and enhances inter-reader consensus in suspected pituitary microadenoma diagnosis. To this end, we developed and validated a stacking model integrating clinical, radiomics, and deep learning features on non-contrast T1COR MRI. This retrospective multicenter study enrolled 636 patients from three centers, divided into training (n = 321), internal validation (n = 138), and two external validation cohorts (n = 136, n = 41). We developed four base models—handcrafted radiomics, deep transfer learning (DTL), deep learning radiomics (DLR), and clinical—and integrated them via a stacking ensemble with logistic regression as the meta-classifier. To evaluate real-world clinical impact, a three-round reader study was conducted with 315 lesions and five radiologists (three juniors, two seniors). Readers assessed non-contrast T1COR MRI unaided, with DTL assistance, and with combined model assistance. Performance metrics included AUC, accuracy, sensitivity, specificity, and inter-reader consensus. The combined model outperformed all single-modality approaches across validation cohorts, achieving AUCs of 0.818 (EVC1) and 0.899 (EVC2) with balanced sensitivity and specificity (EVC2: 0.929 and 0.833, respectively). In the reader study with 315 lesions and five radiologists, AI assistance significantly improved diagnostic accuracy across all experience levels. With combined model assistance, gains were more pronounced: seniors achieved 75.4–79.0
Meningiomas are the most common primary intracranial tumors and are frequently monitored over extended periods. Volumetric assessment typically requires manual segmentation, which is time-consuming and associated with interrater variability. This study aimed to develop and validate a deep learning-based model for the automated segmentation of meningiomas and associated peritumoral edema on preoperative magnetic resonance imaging (MRI). We trained a standard nnU-Net deep learning model on contrast-enhanced T1-weighted and FLAIR MRI scans from 100 patients treated at the University Hospital of Zurich. The model was then externally validated on 88 cases from the meningioma SEG-Class dataset from the Cancer Imaging Archive. Segmentation performance was assessed using the Dice similarity coefficient, Jaccard index, and 95th percentile Hausdorff distance. The model achieved mean Dice scores of 0.87 ± 0.23 for meningioma segmentation and 0.63 ± 0.38 for peritumoral edema in internal cross-validation. On the external validation set, the model achieved scores of 0.86 ± 0.17 for meningioma segmentation and 0.31 ± 0.35 for edema. The deep learning model demonstrated high accuracy in segmenting meningiomas and modest performance for peritumoral edema. These results support the potential utility of automated segmentation tools in clinical workflows. Future work should focus on validating model performance across larger multi-center datasets.
A standardized, population-based three-dimensional (3D) computer-aided design (CAD) model of the intracranial arterial system was developed from time-of-flight magnetic resonance angiography (TOF-MRA) data of the Study of Health in Pomerania (SHIP) cohort, tailored for realistic simulation and experimental neurovascular applications. An averaged intracranial TOF-MRA dataset generated from 4308 individual whole-body MRI examinations of the SHIP cohort was used as the anatomical basis. Intracranial arteries were segmented using 3D Slicer with Frangi-based vessel enhancement, semi-automatic region-growing, and manual refinement to obtain continuous vascular masks. Centerlines were extracted with VMTK (Vascular Modelling Toolkit), and vessel radii were computed via distance mapping. The resulting centerline and radius data were imported into a CAD environment (Creo Parametric) to reconstruct smooth vessel centerlines, generate circular cross-sections, and create a lofted three-dimensional lumen model, which was converted into a hollow geometry with a uniform wall thickness and exported as an STL file for 3D printing. The proposed workflow yielded a geometrically consistent, hollow 3D model of the central intracranial arteries, representing a population-averaged arterial anatomy with smooth vessel courses, gradual diameter transitions, and a closed, continuous wall. The CAD model could be successfully manufactured as a physical 3D-printed phantom and provides a stable, reproducible test environment for digital and in vitro investigations of neurovascular interventions under standardized anatomical conditions. Population-based TOF-MRA data can be transformed into a technically robust and anatomically meaningful intracranial reference model suitable for CAD-based simulation and additive manufacturing. The resulting 3D CAD geometry serves as a reusable reference for comparative studies, methodological validation, early-stage device development, and training in endovascular neurosurgery, without aiming to replace patient-specific models. Time-of-flight magnetic resonance angiography (TOF-MRA)–based vascular modeling. Population-averaged intracranial arterial anatomy. Three-dimensional (3D) cerebrovascular reconstruction. Image-based vessel segmentation and centerline extraction. Computational modeling of cerebral blood vessels.
Identifying new imaging markers may refine risk stratification and guide clinical management for endovascular thrombectomy (EVT). We observed a phenomenon that on a subset of patients’ post-procedural TOF-MRA, the successfully recanalized vessels demonstrated marked vasodilatation in the middle cerebral artery (MCA) territory. The association between global vasodilatation in the MCA territory and clinical outcomes has not yet been investigated. We aimed to characterize the incidence and radiological characteristics of global vasodilatation in the MCA territory, and to investigate its associated risk factors and prognostic implications. Retrospective analysis of 494 consecutive anterior circulation large vessel occlusion stroke (LVOS) patients with successful EVT. In patients with recanalization, global vasodilatation in the MCA territory can be defined if either of the following criteria is met: displayable distal branching number exceeded the contralateral side by > 50
Neurovascular emergencies such as ruptured cerebral aneurysms and acute ischemic stroke (AIS) remain associated with high morbidity and mortality. Stent implantation has become increasingly important in their management, yet these procedures require immediate and reliable platelet inhibition. Cangrelor, an intravenous P2Y12 inhibitor with rapid onset and offset, may offer advantages compared to conventional oral or intravenous regimens. To evaluate the efficacy and safety of cangrelor in patients with acute ischemic stroke (AIS) and in patients with cerebral aneurysms, requiring emergency stent implantation. We retrospectively reviewed 43 consecutive patients with AIS and 24 patients with cervical and cerebral aneurysms treated between January 2021 and March 2026 who received cangrelor during emergency stenting. Indications included AIS with tandem occlusion (n = 35) or intracranial stenosis requiring acute stenting (n = 8), ruptured (n = 17) and acutely symptomatic unruptured (n = 2) aneurysms requiring flow-diverter, stent-assisted coiling (n = 3) or bailout stenting (n = 1) and one extracranial carotid pseudoaneurysm treated with a covered stent. All patients received full dose of cangrelor (bolus 30 µg/kg followed by infusion 4 µg/kg/min for 12–24 h), transitioned to aspirin and ticagrelor. Primary efficacy endpoint was 24-hour stent patency; primary safety endpoint was hemorrhagic complications. Mean age was 66.3 ± 14.1 years for the AIS subgroup and 63.1 ± 16.2 years for the aneurysm subgroup; in AIS cohort 31 patients were male; in cerebral aneurysm 9 patients were male. Hypertension was the most common comorbidity (62.9
To develop a deep learning model for the preoperative, non-invasive identification of germinomas in the pineal region. This retrospective study included 114 patients with pathologically confirmed pineal region tumors. The cohort was randomly divided into a training set (n = 91) and a test set (n = 23). The training set was further partitioned into three folds for cross-validation. A convolutional neural network (CNN) enhanced with contrastive learning was used to extract discriminative features from individual MRI sequences and demographic data. A mixed-attention mechanism was then employed to fuse these features into multimodal representations, aiming to improve identification performance. In this retrospective cohort, the mean age was 9.63 ± 4.99 years, with a male predominance (85.09
