About

Howard A. Riina, MD, is the Lucius N. Littauer Professor of Neurosurgery, Professor of Neurology, and Professor of Radiology at NYU Grossman School of Medicine. He is also the Director of Clinical Innovation at NYU Langone Health and serves as Vice Chair of Clinical Affairs in the Department of Neurosurgery. Dr. Riina specializes in surgical and endovascular treatment for cerebrovascular disorders of the brain and spinal cord, including stroke, brain aneurysms, carotid disease, and vascular malformations. He collaborates with neurologists, neuro-radiologists, and neuro-rehabilitation specialists to treat complex neurovascular conditions, emphasizing a collaborative, patient-centered approach that involves patients and their families in every aspect of care. His research involves developing new devices and treatments for neurovascular conditions, including minimally invasive devices for brain aneurysms, tracheal stents for airway collapse, and intra-arterial chemotherapy for recurrent malignant brain tumors. Dr. Riina has been recognized as one of America’s top doctors by Castle Connolly for over five years and holds leadership roles such as chair of the credentialing endovascular surgery advisory committee and director of the board of directors for The American Board of Neurological Surgery.

Research topics

• Medicine
• Surgery
• Computer Science
• Machine Learning
• Internal medicine
• Artificial Intelligence
• Radiology
• Software engineering
• Pediatrics
• Data science
• Emergency medicine
• Intensive care medicine
• Psychology

Selected publications

• Large Language Models Predict Functional Outcomes after Acute Ischemic Stroke

  arXiv (Cornell University) · 2026-01-18

  articleOpen access

  Accurate prediction of functional outcomes after acute ischemic stroke can inform clinical decision-making and resource allocation. Prior work on modified Rankin Scale (mRS) prediction has relied primarily on structured variables (e.g., age, NIHSS) and conventional machine learning. The ability of large language models (LLMs) to infer future mRS scores directly from routine admission notes remains largely unexplored. We evaluated encoder (BERT, NYUTron) and generative (Llama-3.1-8B, MedGemma-4B) LLMs, in both frozen and fine-tuned settings, for discharge and 90-day mRS prediction using a large, real-world stroke registry. The discharge outcome dataset included 9,485 History and Physical notes and the 90-day outcome dataset included 1,898 notes from the NYU Langone Get With The Guidelines-Stroke registry (2016-2025). Data were temporally split with the most recent 12 months held out for testing. Performance was assessed using exact (7-class) mRS accuracy and binary functional outcome (mRS 0-2 vs. 3-6) accuracy and compared against established structured-data baselines incorporating NIHSS and age. Fine-tuned Llama achieved the highest performance, with 90-day exact mRS accuracy of 33.9% [95% CI, 27.9-39.9%] and binary accuracy of 76.3% [95% CI, 70.7-81.9%]. Discharge performance reached 42.0% [95% CI, 39.0-45.0%] exact accuracy and 75.0% [95% CI, 72.4-77.6%] binary accuracy. For 90-day prediction, Llama performed comparably to structured-data baselines. Fine-tuned LLMs can predict post-stroke functional outcomes from admission notes alone, achieving performance comparable to models requiring structured variable abstraction. Our findings support the development of text-based prognostic tools that integrate seamlessly into clinical workflows without manual data extraction.

  PublisherOA PDF

• 1083 Automated Generation and Human Evaluation of Neurosurgical Board Examination and Self-Assessment Questions

  Neurosurgery · 2026-03-26

  article
  PublisherDOI

• A Multi-AI Agent Framework for Interactive Neurosurgical Education and Evaluation: From Vignettes to Virtual Conversations

  Neurosurgery Open · 2026-03-13

  articleOpen access

  BACKGROUND AND OBJECTIVES: Traditional medical board examinations present clinical information in static vignettes with multiple-choices (MC), fundamentally different from how physicians gather and integrate data in practice. Recent advances in large language models (LLMs) offer promising approaches to creating more realistic clinical interactive conversations. However, these approaches are limited in neurosurgery, where patient communication capacity varies significantly and diagnosis heavily relies on objective data such as imaging and neurological examinations. We aimed to develop and evaluate a multi–artificial intelligence (AI) agent conversation framework for neurosurgical case assessment that enables realistic clinical interactions through simulated patients and structured access to objective clinical data. METHODS: We developed a framework to convert 608 Self-Assessment in Neurological Surgery first-order diagnosis questions into conversation sessions using 3 specialized AI agents: patient AI for subjective information, system AI for objective data, and clinical AI for diagnostic reasoning. We evaluated generative pretrained transformer 4o's (GPT-4o's) diagnostic accuracy across traditional vignettes, patient-only conversations, and patient + system AI interactions, with human benchmark testing from 10 neurosurgery residents. RESULTS: GPT-4o showed significant performance drops from traditional vignettes to conversational formats in both MC (89.0%-60.9%, P < .0001) and free-response scenarios (78.4%-30.3%, P < .0001). Adding access to objective data through system AI improved performance (to 67.4%, P = .0015; and 61.8%, P < .0001, respectively). Questions requiring image interpretation showed similar patterns but lower accuracy. Residents outperformed GPT-4o in free-response conversations (70.0% vs 28.3%, P = .0030) using fewer interactions and reported high educational value of the interactive format. CONCLUSION: This multi-AI agent framework provides both a more challenging evaluation method for LLMs and an engaging educational tool for neurosurgical training. The significant performance drops in conversational formats suggest that traditional MC testing may overestimate LLMs' clinical reasoning capabilities, while the framework's interactive nature offers promising applications for enhancing medical education.

  PublisherDOI

• 2194 Liquid versus Particle Embolic Agents in Preoperative Embolization of Intracranial Meningiomas: A Multicenter Propensity Score–Matched Analysis

  Neurosurgery · 2026-03-26

  article
  PublisherDOI

• Basilar artery perforator rupture as the cause of perimesencephalic subarachnoid hemorrhage.

  Open Access CRIS of the University of Bern · 2026-01-23

  articleOpen access

  Objective The cause of perimesencephalic subarachnoid hemorrhage (pmSAH) is unclear but has historically been attributed to a venous source. The authors hypothesized that high-resolution cone-beam CT (CBCT) during angiography could better identify pmSAH etiology.Methods All patients with pmSAH treated at the authors' institution between January 2023 and December 2024 were retrospectively analyzed. Patients were excluded if CBCT was not performed as part of the digital subtraction angiography (DSA), if CBCT source data were not available for review, or if the images were deemed to be low quality. All images were reviewed by 2 neuroangiographers with extensive neurovascular imaging experience and discussed until consensus agreement. Data were recorded as counts and percentages.Results Among 152 patients who presented with spontaneous SAH in 2023-2024, 22 had a pmSAH defined according to the Rinkel criteria. These 22 patients had a catheter angiogram performed on 1 of 2 biplane machines. Thirteen of those patients had high-quality CBCT data available for review, 8 (61%) of whom were found to harbor a basilar perforator focal outpouching consistent with a site of rupture. All patients with pmSAH, including the 8 found to have a basilar perforator aneurysm, achieved an excellent neurological recovery with resolution of the basilar perforator finding on follow-up DSA with CBCT and without experiencing a re-rupture event or clinically significant vasospasm.Conclusions In the setting of pmSAH, high-resolution CBCT acquired as part of catheter angiography frequently identifies a basilar perforator pseudoaneurysm. Conservative management was associated with excellent outcomes in this series. The authors propose that in the setting of pmSAH, a high suspicion of an arterial etiology should be considered until proven otherwise.

  PublisherDOI

• 2023 Flow Diversion in Fibromuscular Dysplasia: A Decade of Experience

  Neurosurgery · 2026-03-26

  article
  PublisherDOI

• Basilar artery perforator rupture as the cause of perimesencephalic subarachnoid hemorrhage

  Journal of neurosurgery · 2026-01-01

  article

  OBJECTIVE: The cause of perimesencephalic subarachnoid hemorrhage (pmSAH) is unclear but has historically been attributed to a venous source. The authors hypothesized that high-resolution cone-beam CT (CBCT) during angiography could better identify pmSAH etiology. METHODS: All patients with pmSAH treated at the authors' institution between January 2023 and December 2024 were retrospectively analyzed. Patients were excluded if CBCT was not performed as part of the digital subtraction angiography (DSA), if CBCT source data were not available for review, or if the images were deemed to be low quality. All images were reviewed by 2 neuroangiographers with extensive neurovascular imaging experience and discussed until consensus agreement. Data were recorded as counts and percentages. RESULTS: Among 152 patients who presented with spontaneous SAH in 2023-2024, 22 had a pmSAH defined according to the Rinkel criteria. These 22 patients had a catheter angiogram performed on 1 of 2 biplane machines. Thirteen of those patients had high-quality CBCT data available for review, 8 (61%) of whom were found to harbor a basilar perforator focal outpouching consistent with a site of rupture. All patients with pmSAH, including the 8 found to have a basilar perforator aneurysm, achieved an excellent neurological recovery with resolution of the basilar perforator finding on follow-up DSA with CBCT and without experiencing a re-rupture event or clinically significant vasospasm. CONCLUSIONS: In the setting of pmSAH, high-resolution CBCT acquired as part of catheter angiography frequently identifies a basilar perforator pseudoaneurysm. Conservative management was associated with excellent outcomes in this series. The authors propose that in the setting of pmSAH, a high suspicion of an arterial etiology should be considered until proven otherwise.

  PublisherDOI

• Safety of endovascular shunting for normal pressure hydrocephalus from a prospective, multicenter, single-arm study

  Journal of NeuroInterventional Surgery · 2026-02-24

  articleOpen access

  OBJECTIVE: Normal pressure hydrocephalus (NPH) is a neurological condition characterized by impaired gait, cognitive decline, and urinary incontinence. Endovascular shunting using the eShunt Implant diverts cerebrospinal fluid (CSF) from basal cisterns to the internal jugular vein via a transvenous, transfemoral procedure. We report initial eShunt safety data for treatment of NPH through 90 days in a prospective, multicenter, single-arm trial. METHODS: NPH participants indicated for endovascular shunt placement were included after exhibiting positive gait response to lumbar CSF withdrawal and candidate anatomy. NPH symptoms at baseline and following eShunt Implant placement were assessed using the Timed Up & Go (TUG) test for gait, Montreal Cognitive Assessment (MoCA) for cognition, and Neurogenic Bladder Symptom Score-Short Form (NBSS-SF) for urinary symptoms. The primary endpoint was the rate of device and/or procedure-related serious adverse events (SAE) at 90-day follow-up. RESULTS: Sixty-six participants were treated with the eShunt Implant without immediate or delayed cerebral hemorrhage, over-drainage, infection or device-related SAEs. Two procedure-related SAEs (3.0%) occurred, one sigmoid sinus thrombosis and one femoral artery pseudoaneurysm, both resolved without surgical intervention. Matched-pair analysis showed statistically significant improvements in TUG time (37.2%; P<0.0001), MoCA score (+2.3; P<0.0001), and NBSS-SF score (-2.1; P<0.002). CONCLUSIONS: Endovascular shunting for NPH was feasible with a low rate of SAEs typically associated with conventional shunt surgery. An acceptable clinical response through 90 days was observed. These findings suggest further investigation of this minimally invasive, endovascular approach to NPH in a randomized controlled trial comparing to standard of care. TRIAL REGISTRATION NUMBER: ClinicalTrials.gov Identifiers: NCT05250505 and NCT05232838.

  PublisherOA PDFDOI

• Large Language Models Predict Functional Outcomes after Acute Ischemic Stroke

  Open MIND · 2026-01-18

  preprint

  Accurate prediction of functional outcomes after acute ischemic stroke can inform clinical decision-making and resource allocation. Prior work on modified Rankin Scale (mRS) prediction has relied primarily on structured variables (e.g., age, NIHSS) and conventional machine learning. The ability of large language models (LLMs) to infer future mRS scores directly from routine admission notes remains largely unexplored. We evaluated encoder (BERT, NYUTron) and generative (Llama-3.1-8B, MedGemma-4B) LLMs, in both frozen and fine-tuned settings, for discharge and 90-day mRS prediction using a large, real-world stroke registry. The discharge outcome dataset included 9,485 History and Physical notes and the 90-day outcome dataset included 1,898 notes from the NYU Langone Get With The Guidelines-Stroke registry (2016-2025). Data were temporally split with the most recent 12 months held out for testing. Performance was assessed using exact (7-class) mRS accuracy and binary functional outcome (mRS 0-2 vs. 3-6) accuracy and compared against established structured-data baselines incorporating NIHSS and age. Fine-tuned Llama achieved the highest performance, with 90-day exact mRS accuracy of 33.9% [95% CI, 27.9-39.9%] and binary accuracy of 76.3% [95% CI, 70.7-81.9%]. Discharge performance reached 42.0% [95% CI, 39.0-45.0%] exact accuracy and 75.0% [95% CI, 72.4-77.6%] binary accuracy. For 90-day prediction, Llama performed comparably to structured-data baselines. Fine-tuned LLMs can predict post-stroke functional outcomes from admission notes alone, achieving performance comparable to models requiring structured variable abstraction. Our findings support the development of text-based prognostic tools that integrate seamlessly into clinical workflows without manual data extraction.

  DOI

• 1082 Tabular Foundation Models Predict Clinical Outcomes in Skull Base and Cerebrovascular Surgery

  Neurosurgery · 2026-03-26

  article
  PublisherDOI

Frequent coauthors

• Y. Pierre Gobin

  Cornell University

  189 shared

• Philip E. Stieg

  Cornell University

  146 shared

• Alejandro Santillán

  The University of Texas at San Antonio

  120 shared

• Jeffrey M. Katz

  116 shared

• Apostolos John Tsiouris

  Presbyterian Hospital

  111 shared

• Athos Patsalides

  North Shore University Hospital

  97 shared

• Eytan Raz

  NYU Langone Health

  85 shared

• John A. Boockvar

  Lenox Hill Hospital

  84 shared

Awards & honors

• Named one of America’s top doctors by Castle Connolly for mo…