Faculty Bibliography

BACKGROUND AND OBJECTIVES/OBJECTIVE:General purpose vision-language models (VLMs) demonstrate impressive capabilities, but their opaque training on uncurated internet data poses critical limitations for high-stakes decision making, such as in neurosurgery. We present CNS-Obsidian, a neurosurgical VLM trained on peer-reviewed neurosurgical literature, and demonstrate its clinical utility compared with GPT-4o in a real-world setting. METHODS:We compiled 23 984 articles from Neurosurgery Publications journals, yielding 78 853 figures and captions. Using GPT-4o and Claude Sonnet-3.5, we converted these image-text pairs into 263 064 training samples across 3 formats: instruction fine-tuning, multiple-choice questions, and differential diagnosis. We trained CNS-Obsidian, a fine-tune of the 34-billion parameter Large Language and Visual Assistant-Next model. In a blinded, randomized deployment trial at NYU Langone Health (August 30-November 30, 2024), neurosurgeons were assigned to use either CNS-Obsidian or a Health Insurance Portability and Accountability Act-compliant GPT-4o end point as a diagnostic copilot after patient consultations. Primary outcomes were diagnostic helpfulness and accuracy, assessed through user ratings and presence of the correct diagnosis within the VLM-provided differential, respectively. RESULTS:CNS-Obsidian matched GPT-4o on synthetic questions (76.13% vs 77.54%, P = .235), but only achieved 46.81% accuracy on human-generated questions vs GPT-4o's 65.70% (P < 10-15). In the randomized trial, 70 consultations were evaluated (32 CNS-Obsidian, 38 GPT-4o) from 959 total consults (7.3% utilization). CNS-Obsidian received positive ratings in 40.62% of cases vs 57.89% for GPT-4o (P = .230). Both models included correct diagnosis in approximately 60% of cases (59.38% vs 65.79%, P = .626). CONCLUSION/CONCLUSIONS:Domain-specific VLMs trained on curated scientific literature can approach frontier model performance in specialized medical domains despite being orders of magnitude smaller and less expensive to train. This establishes a transparent framework for scientific communities to build specialized artificial intelligence models. However, low clinical utilization suggests chatbot interfaces may not align with specialist workflows, indicating need for alternative artificial intelligence integration strategies.

Episignatures are increasingly valuable for variant interpretation in rare neurodevelopmental disorders, especially when optimized to capture the impact of specific variant types and locations across a gene. Here, we generated a next-generation DNA methylation (DNAm) episignature for Kabuki syndrome type 1 (KS1) using the largest cohort studied to date, aiming to clarify the epigenomic and phenotypic effects of diverse KMT2D variant types and positions. Genome-wide DNAm profiles were obtained for 110 individuals with KMT2D variants and 854 controls using microarrays and long-read sequencing (LRS). Differentially methylated loci were enriched in genes involved in embryonic and nervous system development and were leveraged to construct a support-vector machine classifier for detecting pathogenic KMT2D variants. The classifier achieved 97% sensitivity and 100% specificity in validation cohorts and outperformed in silico tools, demonstrating stronger concordance with clinical presentation. Missense variants in the C-terminal region (exon 48) of KMT2D and the N-terminal plant homeodomain (PHD)-type zinc fingers were predominantly classified as pathogenic, highlighting regions enriched for pathogenic variants. Missense variants in the central region (exons 31-39) were more often predicted benign for KS1, consistent with potential association with a different syndrome, highlighting the classifier's specificity for KS1. Test performance was consistent across array and LRS platforms, and classifier scores reflected levels of mosaicism detected by LRS. The KS1 episignature also positively classified pathogenic KDM6A variants associated with KS2. These findings represent a significant advance in the evolution of episignature development, demonstrating diagnostic and interpretive value of a KS1 signature in resolving uncertain or complex cases.

BACKGROUND AND OBJECTIVES/OBJECTIVE:articles into graphical abstracts using Cascade Styling Sheets (CSS) templates and iterative prompting of a frontier vision language model and to conduct a human evaluation of this pipeline. METHODS:We developed an automated pipeline to convert extracted manuscript content into standardized graphical abstracts. The pipeline implements a custom CSS profile designed to match existing journal standards. Using Claude Sonnet-3.5, we generated structured hypertext markup language summaries organized into 6 sections: Objectives, Background, Methods, Results, Discussion, and Conclusion. The model selected up to 2 representative figures per manuscript based on caption analysis. We evaluated performance using 100 randomly selected articles published between 2020 and 2024 (95 from Neurosurgery, 4 from Operative Neurosurgery, 1 from Neurosurgery Practice). Three Editorial Review Board members independently assessed abstracts using 3 binary criteria: (1) proper formatting, (2) factual accuracy, and (3) visual appeal. RESULTS:Generated graphical abstracts achieved proper formatting in 85% of cases (95% CI: 76.7%-90.7%), factual accuracy in 99% (95% CI: 94.4%-99.9%), and visual appropriateness in 82% (95% CI: 73.3%-88.3%). Overall, 70% of abstracts (95% CI: 60.5%-78.1%) met all 3 criteria and were deemed "publication ready" without manual intervention. Error analysis revealed poor figure selection (40.0%) as the most common failure mode, followed by title replacement errors from PDF extraction (26.7%). CONCLUSION/CONCLUSIONS:Our artificial intelligence-CSS pipeline demonstrates the feasibility of automating graphical abstract generation for neurosurgical manuscripts, achieving publication-ready quality in 70% of cases with 99% factual accuracy. This technology offers a scalable augmentation tool that can reduce the design burden for authors, enhancing visual scientific communication in neurosurgical publishing while complementing human expertise.

Rasmussen syndrome (RS) includes a well-described constellation of refractory focal seizures, often including epilepsia partialis continua, hemiplegia with progressive unilateral cortical atrophy, and cognitive/language decline. However, the precise early pathogenesis and reliable biomarkers remain elusive. In addition, we lack operational management guidelines, including diagnostic evaluation, disease-monitoring assessments, and medical and surgical treatment approaches. We aimed to create an expert consensus statement to guide and standardize the treatment of RS, with the goal of providing recommendations applicable to a global population. An expert panel was convened to complete three rounds of a modified Delphi procedure given the lack of high-level evidence, with a focus on workup to exclude mimicking diagnoses, disease-activity metrics, and treatment. Consensus was defined as ≥75% of responses being agree/strongly agree in either two subsequent rounds or in the third and final round. A total of 122 of 143 statements met consensus. Proposed diagnostic evaluation in patients with possible RS is outlined, including physical examination, blood/cerebrospinal fluid analyses, neuroimaging, electroencephalography (EEG), and biopsy. Suggested disease-monitoring assessments include neuropsychological testing and serial magnetic resonance imaging (MRI). Intravenous corticosteroids are recommended as first-line, acute immunotherapy for seizure exacerbations and status epilepticus, with or without the addition of intravenous immunoglobulin. Options for maintenance immunotherapy are outlined, with lack of evidence noted for comparing efficacy of these treatments. Hemispheric disconnection remains the most effective seizure treatment, with parameters including age, function, seizure burden, and patient values influencing candidacy for surgery. This consensus statement offers a guideline to standardize management, as well as suggests future directions to further elucidate underlying pathophysiology and target more-effective, better-tolerated treatments.

Neuronal axons have traditionally been considered to be the primary mediators of functional connectivity among brain regions. However, the role of astrocyte-mediated communication has been largely underappreciated. Astrocytes communicate with one another through gap junctions, but the extent and specificity of this communication remain poorly understood. Astrocyte gap junctions are necessary for memory formation^1,2, synaptic plasticity^3-5, coordination of neuronal signalling⁶, and closing the visual and motor critical periods^7,8. These findings indicate that this form of communication is essential for proper central nervous system development and function. Despite the importance of astrocyte gap junctional networks, studying them has been challenging. Current methods such as slice electrophysiology disrupt network connectivity and introduce artefacts due to tissue damage. Here, we developed a vector-based approach that labels molecules as they are fluxed by astrocyte gap junctions in awake, behaving animals to overcome these limitations. We then used whole-brain tissue clearing^9,10 to image these intact, three-dimensional astrocyte networks. We show that multiple astrocyte networks traverse the mouse brain. These networks selectively connect specific regions, rather than diffusing indiscriminately, and vary in size and organization. We observe local networks that are confined to single brain regions and long-range networks that robustly interconnect multiple regions across hemispheres, often exhibiting patterns distinct from known neuronal networks. We also demonstrate that astrocyte networks undergo structural reorganization in the adult brain after sensory deprivation. These findings reveal a mode of communication between distant brain regions that is mediated by plastic networks of gap junction-coupled astrocytes.

Head computed tomography (CT) imaging is a widely used imaging modality with multitudes of medical indications, particularly in assessing pathology of the brain, skull and cerebrovascular system. It is commonly used as the first-line imaging in neurologic emergencies given its rapidity of image acquisition, safety, cost and ubiquity. Deep learning models may facilitate detection of a wide range of diseases. However, the scarcity of high-quality labels and annotations, particularly among less common conditions, substantially hinders the development of powerful models. To address this challenge, we introduce FM-HCT, a Foundation Model for Head CT for generalizable disease detection, trained using self-supervised learning. Our approach pretrains a deep learning model on a large, diverse dataset of 361,663 non-contrast 3D head CT scans without the need for manual annotations, enabling the model to learn robust, generalizable features. Our results demonstrate that the self-supervised foundation model substantially improves performance on downstream diagnostic tasks compared to models trained from scratch and previous 3D CT foundation models trained on scarce annotated datasets.

BACKGROUND:Transcranial electrical stimulation (TES) has limited spatial focus and depth penetration, constraining its therapeutic efficacy. Intersectional Short-Pulse (ISP) stimulation was developed to overcome these limitations by delivering rapidly switching pulses that can be temporally integrated by neuronal membranes. Here, we aimed to establish the biophysical basis of ISP-induced temporal summation and to test whether this mechanism enables effective brain modulation in vivo. METHODS:We combined finite-element modeling, cadaver measurements (n = 2 human cadavers), and biophysically realistic NEURON simulations to characterize the spatial and temporal properties of ISP-induced electric fields. In vivo whole-cell patch-clamp recordings were performed in the rat somatosensory cortex (female Wistar rat) to test the membrane-level integration of sequential electric field pulses. Functional efficacy was evaluated using closed-loop ISP stimulation in a hippocampal kindling model of temporal lobe epilepsy in male Long-Evans rats (n = 11 animals, >500 induced seizures analyzed across conditions). RESULTS:Here we show that neurons integrate sequential ISP pulses in a non-vectorial, temporally accumulative manner, consistent with membrane-level charge integration rather than extracellular field superposition. ISP and conventional TES simulations produced similar instantaneous field magnitudes, but ISP stimulation resulted in more uniform neuronal excitability across brain depths. Closed-loop ISP stimulation significantly outperformed conventional TES in reducing seizure duration and severity. ISP reduced hippocampal seizure duration by 45% and 35% compared to SHAM stimulation and conventional TES, and significantly reduced motor seizure severity. CONCLUSIONS:ISP stimulation provides a non-invasive neuromodulation approach that enhances deep brain engagement through rapid, temporally structured pulse sequences. These findings demonstrate effective seizure suppression in a rodent model and support the translational potential of ISP for disorders involving pathological neural dynamics.

INTRODUCTION/BACKGROUND:Measuring plasma biomarkers effectively assesses early-stage Alzheimer's disease. METHODS:Subjects were categorized as cognitively unimpaired (CU) (n = 66), CU with subjective cognitive decline (SCD) (n = 100), and mild cognitive impairment (MCI) (n = 25). Plasma biomarkers measured were amyloid beta (Aβ) 40, Aβ42, neurofilament light chain (NfL), glial fibrillary acidic protein (GFAP), tau phosphorylated at threonine 181 (pTau181), neuroinflammatory biomarkers, and blood-brain barrier biomarkers. Amyloid and tau positron emission tomography (PET) imaging was performed in 186 and 144 subjects, respectively. RESULTS:Comparing those having MCI, both CU and SCD participants had significantly lower amyloid PET standardized uptake value ratio (SUVR) (p < 0.001; p = 0.005). Higher amyloid PET SUVR was significantly associated with higher pTau181 (p = 0.001) and a higher pTau181/Aβ42 ratio (p < 0.001). Higher tau PET SUVR was associated with lower plasma Aβ42 (p = 0.020), older age (p = 0.005), higher GFAP (p = 0.020), and lower interleukin-8 levels (p < 0.001). DISCUSSION/CONCLUSIONS:Our study supports plasma biomarker monitoring of at-risk patients at various stages of pre-dementia.

IMPORTANCE/UNASSIGNED:Seizure-induced brain injury is central to the treatment urgency of new-onset refractory status epilepticus (NORSE). Identifying biomarkers that reflect ongoing neuronal damage could inform therapeutic timing and improve outcomes. OBJECTIVE/UNASSIGNED:To quantify acute brain injury in patients with cryptogenic NORSE (cNORSE), etiology-defined status epilepticus (eSE), and chronic epilepsy. DESIGN, SETTING, AND PARTICIPANTS/UNASSIGNED:This was an international cross-sectional study conducted between 2013 and 2025. Patients were enrolled at 36 hospitals in the US, 2 in Canada, and 1 in Italy, France, and Belgium. Patients with cNORSE and eSE for which biological samples were obtained during ongoing seizure activity were enrolled in the study. Comparison groups without status epilepticus comprised individuals with chronic epilepsy and healthy participants. None were excluded. EXPOSURES/UNASSIGNED:Neurofilament light chain (NfL) and S100-beta (S100B) protein concentrations in serum and cerebrospinal fluid (CSF). MAIN OUTCOMES AND MEASURES/UNASSIGNED:Degree of neuronal and glial damage, indexed by NfL and S100B levels, and their association with short-term functional outcomes. RESULTS/UNASSIGNED:A total of 78 patients with cNORSE (mean [95% CI] age, 37 [30-41] years; 44 female [56%]) and 2 independent cohorts of 211 patients (mean [95% CI] age, 69 [66-71] years; 128 female [61%]) and 73 patients (mean [95% CI] age, 56 [45-65] years; 39 male [53%]) with eSE were included. NfL concentrations were markedly elevated in cNORSE-approximately 10-fold higher in CSF and 4-fold higher in serum-compared with the eSE cohorts (CSF: median [IQR], 6408 [1503-22 963] pg/mL compared with 694 [219-2389] pg/mL; serum: median [IQR], 231 [99-855] pg/mL compared with 55 [20-135] pg/mL; P <.001). Serum NfL levels were nearly 20-fold higher in cNORSE than in the cohort with epilepsy and in healthy controls (median [IQR], 11 [7-19 ] and 7 [5-14 ] pg/mL, respectively). Serum and CSF NfL levels were strongly correlated (Spearman ρ = 0.75; P < .001) and rose sharply between week 1 (median [IQR], 101 [51-137] pg/mL), week 2 (median [IQR], 197 [117-324] pg/mL), and week 3 (median [IQR], 598 [163-1000] pg/mL) after onset (P < .001). In contrast, S100B concentrations did not differ between groups and showed no consistent temporal pattern. NfL discriminated cNORSE from eSE (area under the receiver operating characteristic curve [AUROC], 0.79; 95% CI, 0.68-0.90) and from cohorts without status epilepticus (AUROC, 0.99; 95% CI, 0.78-1.00). Higher serum NfL was independently associated with poor functional outcome at discharge (Glasgow Outcome Scale extended score, 1-4; odds ratio, 1.01; 95% CI, 1.00-1.03; P = .03). CONCLUSIONS AND RELEVANCE/UNASSIGNED:Results of this cross-sectional study suggest that acute neuroaxonal injury, as reflected by elevated NfL levels, was substantially greater in cNORSE than in the cohorts with eSE and in controls without status epilepticus. The rapid early rise in NfL highlights a narrow therapeutic window, emphasizing the need for prompt, effective, and potentially neuroprotective interventions in cNORSE.

Ceramide transporter syndrome (CerTra syndrome) is a rare neurodevelopmental disorder caused by pathogenic variants in CERT1 gene encoding ceramide transporter (CERT). These variants disrupt ceramide transport and sphingolipid homeostasis, leading to a clinical phenotype that includes developmental delay, movement abnormalities, and structural brain anomalies. Despite growing recognition of this condition, detailed neuroimaging and neuropathological characterization remain limited. Here, we present a 12-year-old girl with a pathogenic CERT1 variant complicated by sudden unexplained death, who presented with unreported neuroimaging abnormalities in choroid plexus (ChP) and perivascular space (PVS). Clinical Magnetic Resonance Imaging (MRI) obtained approximately 10 years prior to death, as well as postmortem MRI, revealed bilateral cystic enlargement of the ChP and prominent PVS filled with abundant proteinaceous material in white matter. Neuropathological examination demonstrated marked ChP epithelial disorganization, reduced aquaporin-1 (AQP1) expression, cyst formation, and focal calcifications, which may be associated with disturbances in cerebrospinal fluid (CSF) dynamics. These findings raise the possibility that CERT1 variants may be associated with ChP architectural changes and altered perivascular clearance.