2026 Poster Presentations
P486: ANATOMICAL PHENOTYPING AND STAGING OF BRAIN ARTERIOVENOUS MALFORMATIONS
Benjamin Beyersdorf, MD1; Yannis Schwieger, MD2; Luis Padevit, MD1; Zsolt Kulcsar3; Kunal Vakharia, MD4; Christopher S Ogilvy, MD5; Harry van Loveren, MD4; Menno Germans, MD, PhD1; Philipp Taussky, MD5; Siviero Agazzi, MD4; Luca Regli, MD1; Kevin Akeret, MD, PhD6; 1Department of Neurosurgery, Clinical Neuroscience Centre, University Hospital Zurich and University of Zurich, Zurich, Switzerland; 2Department of Surgery, Stadtspital Triemli, Zurich, Switzerland; 3Department of Neuroradiology, Clinical Neuroscience Centre, University Hospital Zurich and University of Zurich, Zurich, Switzerland; 4Department of Neurosurgery and Brain Repair, University of South Florida, Morsani College of Medicine, Tampa, Florida, United States; 5Department of Neurosurgery, Beth Israel Deaconess, Harvard Medical School, Boston, Massachusetts; 6Department of Neurosurgery and Brain Repair, University of South Florida, Morsani College of Medicine, Tampa, Florida, United States & Department of Neurosurgery, Clinical Neuroscience Centre, University Hospital Zurich and University of Zurich, Zurich, Switzerland
Brain arteriovenous malformations (AVMs) are potentially life-threatening vascular anomalies that pose significant clinical challenges due to their heterogeneous anatomy and unpredictable natural history. Existing risk stratification models largely rely on isolated imaging markers and fail to account for the dynamic spatial-temporal complexity of AVMs. Building on our prior work, demonstrating how ontogenesis dictates clinical outcomes of neuroepithelial tumors, we hypothesize that AVMs are similarly influenced by developmental processes that define their spatial distribution, vascular architecture, and susceptibility to complications. Here, we present the protocol and pilot data of our multicenter, retrospective and prospective observational study, which introduces a multidimensional approach integrating precise anatomical phenotyping with ontogenetic mapping and the analysis of dynamic structural changes over time. By leveraging unsupervised non-negative matrix factorization, we aim to identify higher-order AVM meta-topologies that link structural and developmental patterns with clinical events. Ultimately, the objective of this integrative framework is to facilitate the development of a robust biologically informed risk-stratifying AVM staging system, enhancing personalized treatment strategies and optimizing patient outcomes.
Fig. 1: Spatiotemporal Ontogenetic Identity in Brain Arteriovenous Malformations. The upper part of this figure illustrates the physiological ontogenesis from conception to the postnatal brain, depicted for parenchymal, arterial, and venous units. It emphasizes different hypothesized ontogenetic inflection points, marking the onset of deviation from normal development, which may influence the AVM's parenchymal distribution, vascular architecture, and clinical behavior.
Fig. 2: The Multidimensional Spatiotemporal Approach. The top row illustrates the natural course of AVM evolution, the lower section outlines the study workflow: three-dimensional anatomical phenotyping of mature AVMs is performed and additional dynamic structural features are dissected. Next, anatomical phenotypes are integrated to identify AVM meta-topologies and decipher their corresponding ontogenetic identity. Clinical risk assessment combines meta-topologies and dynamic structural features to inform clinical risk stratification and support the development of an anatomical staging system.
Fig. 3: Topographic Probability of Brain AVMs. Graphical representation of AVM topographic probability across parenchymal structures. The heatmap (left) illustrates the probability of AVM occurrence in each anatomical unit after normalization for structure volume, while the anatomical illustration (left) overlays this probability distribution onto lateral and medial brain views.
Fig. 4: Deconvolution of Meta-Topologies. Graphical representation of six deconvoluted meta-topologies. Each panel consists of a heatmap overlaid on a lateral (left) and medial (right) brain scaffold.