About

Xianfeng Gu is a SUNY Empire Innovation Professor in the Department of Computer Science at Stony Brook University. He received his Ph.D. in Computer Science from Harvard University in 2003 and his B.S. from Tsinghua University in Beijing, China, in 1995. His research focuses on applying modern geometry in engineering and medical fields. He systematically develops discrete theories and computational algorithms within the interdisciplinary field of Computational Conformal Geometry, applying them to solve real-world problems such as global surface parameterization in graphics, deformable shape registration in vision, manifold spline in geometric modeling, curvature convergence analysis in geometric processing, efficient routing in networking, brain mapping, and virtual colonoscopy in medical imaging. Gu has received several awards, including the Morningside Gold Medal of Applied Mathematics at the International Congress of Chinese Mathematicians in 2013, the Research Excellence Award from the Stony Brook University Department of Computer Science in 2010, the Gaheon Award for the best paper in the International Journal of CAD/CAM in 2009, and the NSF Faculty Early Career Award in 2005.

Research topics

• Artificial Intelligence
• Computer Science
• Machine Learning
• Mathematics
• Algorithm
• Structural engineering
• Physics
• Geometry
• Engineering

Selected publications

• Aneurysm Morphology Based on Conformal Geometry

  International Journal for Numerical Methods in Biomedical Engineering · 2026-05-01

  articleSenior authorCorresponding

  Several morphological parameters such as aortic neck length, angulation, or centerline curvature have previously been evaluated to define "hostile" anatomies that predispose to poor outcomes after endovascular aneurysm repair. In this study, we present a new method for classifying aneurysm morphologies using their conformal structures. The conformal structure of a surface is determined by its Riemannian metric, and conformal mappings between surfaces with complex topologies preserve their conformal structures. To classify aneurysm shapes, the aortic aneurysm is first segmented from CT scans and represented as a discrete surface. Next, holomorphic differential forms are computed based on the discrete Hodge Theory. The aneurysm surface can be conformally mapped onto a pair of planar rectangles with an aortic bifurcation point by integrating a special holomorphic differential. This canonical configuration gives the conformal invariants, or "conformal fingerprints," which can then be used to classify aneurysm morphologies. This novel methodology shows promise in providing an improved understanding of aneurysm morphologies, which can aid in better predicting and managing potential complications after endovascular aneurysm repair.

  PublisherDOI

• Inverse Rendering for High-Genus 3D Surface Meshes from Multi-View Images with Persistent Homology Priors

  2026-04-21

  articleSenior author

  Reconstructing 3D objects from images is inherently an ill-posed problem due to ambiguities in geometry, appearance, and topology. This paper introduces collaborative inverse rendering with persistent homology priors, a novel strategy that leverages topological constraints to resolve these ambiguities. By incorporating priors that capture critical features such as tunnel loops and handle loops, our approach directly addresses the difficulty of reconstructing high-genus surfaces. The collaboration between photometric consistency from multi-view images and homology-based guidance enables recovery of complex high-genus geometry while circumventing catastrophic failures such as collapsing tunnels or losing high-genus structure. Instead of neural networks, our method relies on gradient-based optimization within a mesh-based inverse rendering framework to highlight the role of topological priors. Experimental results show that incorporating persistent homology priors leads to lower Chamfer Distance (CD) and higher Volume IoU compared to state-of-the-art mesh-based methods, demonstrating improved geometric accuracy and robustness against topological failure.

  PublisherDOI

• Multiple data sets to explore the key molecules and mechanism of lymph node metastasis in gastric cancer

  Discover Oncology · 2025-04-24 · 1 citations

  articleOpen access

  OBJECTIVE: To explore the key molecules and regulatory mechanisms of lymph node metastasis in gastric cancer. METHODS: The differential genes and key genes of lymph node metastasis in gastric cancer were analyzed by utilizing multiple data sets. The key genes were analyzed by GSEA analysis, transcription factor analysis, nomogram prediction model construction, immune infiltration analysis, GSVA analysis, drug sensitive analysis and single cell data analysis. RESULTS: Abnormal expression of key genes including CDRT15P1, DENND3, F2R, FNDC3B, IRAK3, MS4A2, PDK4, PKIA and activation of related signaling pathways might be the result of ultraviolet radiation-induced DNA damage, which was closely related to lymph node metastasis in gastric cancer. The key genes were regulated by a variety of transcription factors, which were strongly connected with the invasion of immune cells and the sensitivity of a variety of drugs. The nomogram prediction model, which is based on the key genes associated with lymph node metastasis and the TNM of gastric cancer, demonstrated a high level of predictive efficiency. CONCLUSION: CDRT15P1, DENND3, F2R, FNDC3B, IRAK3, MS4A2, PDK4 and PKIA may be the key genes affecting lymph node metastasis in gastric cancer, and F2R has higher biological importance.

  PublisherOA PDFDOI

• OccludeNeRF: Geometric-aware 3D Scene Inpainting with Collaborative Score Distillation in NeRF

  ArXiv.org · 2025-04-01

  preprintOpen access

  With Neural Radiance Fields (NeRFs) arising as a powerful 3D representation, research has investigated its various downstream tasks, including inpainting NeRFs with 2D images. Despite successful efforts addressing the view consistency and geometry quality, prior methods yet suffer from occlusion in NeRF inpainting tasks, where 2D prior is severely limited in forming a faithful reconstruction of the scene to inpaint. To address this, we propose a novel approach that enables cross-view information sharing during knowledge distillation from a diffusion model, effectively propagating occluded information across limited views. Additionally, to align the distillation direction across multiple sampled views, we apply a grid-based denoising strategy and incorporate additional rendered views to enhance cross-view consistency. To assess our approach's capability of handling occlusion cases, we construct a dataset consisting of challenging scenes with severe occlusion, in addition to existing datasets. Compared with baseline methods, our method demonstrates better performance in cross-view consistency and faithfulness in reconstruction, while preserving high rendering quality and fidelity.

  PublisherOA PDFDOI

• Nano-enabled repurposing of Benproperine phosphate enhances pancreatic Cancer chemotherapy through lethal autophagy arrest and immune activation

  Journal of Controlled Release · 2025-09-01 · 3 citations

  article
  PublisherDOI

• Hyper-Spherical Optimal Transport for Semantic Alignment in Text-to-3D End-to-End Generation

  IEEE Transactions on Visualization and Computer Graphics · 2025-07-07

  articleSenior author

  Recent CLIP-guided 3D generation methods have achieved promising results but struggle with generating faithful 3D shapes that conform with input text due to the gap between text and image embeddings. To this end, this paper proposes HOTS3D which makes the first attempt to effectively bridge this gap by aligning text features to the image features with spherical optimal transport (SOT). However, in high-dimensional situations, solving the SOT remains a challenge. To obtain the SOT map for high-dimensional features obtained from CLIP encoding of two modalities, we mathematically formulate and derive the solution based on Villani's theorem, which can directly align two hyper-sphere distributions without manifold exponential maps. Furthermore, we implement it by leveraging input convex neural networks (ICNNs) for the optimal Kantorovich potential. With the optimally mapped features, a diffusion-based generator is utilized to decode them into 3D shapes. Extensive quantitative and qualitative comparisons with state-of-the-art methods demonstrate the superiority of HOTS3D for text-to-3D generation, especially in the consistency with text semantics.

  PublisherDOI

• 536 Topology optimization for personalized intracranial aneurysm implant design

  Journal of Clinical and Translational Science · 2025-03-25

  articleOpen accessSenior author

  Objectives/Goals: To develop a personalized computational framework integrating computational fluid dynamics (CFD) and topology optimization for designing intracranial aneurysm implants. The primary objective is to reduce intra-aneurysmal blood flow velocity and enhance thrombus formation for improved treatment outcomes. Methods/Study Population: Patient-specific aneurysm geometries were extracted from pre-treatment rotational angiograms. A CFD-driven topology optimization framework was employed to design implants that reduce intra-aneurysmal flow velocity. The fluid dynamics were modeled using Navier–Stokes equations and the structural integrity of the implants was ensured by linear elasticity equations. The solid isotropic material with penalization (SIMP) method was applied to optimize the implant’s porous architecture, balancing flow reduction with structural support. COMSOL Multiphysics software was used to implement the optimization. Results/Anticipated Results: The optimized implants demonstrated significant reductions in intra-aneurysmal blood flow velocity and improved hemodynamic conditions. Flow velocity within the aneurysm was reduced by 77%, and the fluid energy dissipation ratio showed a 78.9% improvement compared to pretreatment conditions. The optimized porous structures were tailored to the aneurysm’s specific geometry, providing personalized designs that improve flow stasis and thrombus formation. Further validation of the implants will be performed in vitro and in vivo to assess their effectiveness and biocompatibility. Discussion/Significance of Impact: This personalized implant design framework could lead to better treatment outcomes by reducing aneurysm recurrence and complications compared to current devices. It provides a pathway for improved occlusion rates and patient-specific solutions for intracranial aneurysms.

  PublisherOA PDFDOI

• Brain age prediction based on Aβ-PET and random forest in ADNI: modulation by APOE4 genotype and SHAP interpretation

  2025-12-04

  article
  PublisherDOI

• OT-Talk: Animating 3D Talking Head with Optimal Transportation

  2025-06-25 · 2 citations

  articleSenior author

  Animating 3D head meshes using audio inputs has significant applications in AR/VR, gaming, and entertainment through 3D avatars. However, bridging the modality gap between speech signals and facial dynamics remains a challenge, often resulting in incorrect lip syncing and unnatural facial movements. To address this, we propose OT-Talk, the first approach to leverage optimal transportation to optimize the learning model in talking head animation. Building on existing learning frameworks, we utilize a pre-trained Hubert model to extract audio features and a transformer model to process temporal sequences. Unlike previous methods that focus solely on vertex coordinates or displacements, we introduce Chebyshev Graph Convolution to extract geometric features from triangulated meshes. To measure mesh dissimilarities, we go beyond traditional mesh reconstruction errors and velocity differences between adjacent frames. Instead, we represent meshes as probability measures and approximate their surfaces. This allows us to leverage the sliced Wasserstein distance for modeling mesh variations. This approach facilitates the learning of smooth and accurate facial motions, resulting in coherent and natural facial animations. Our experiments on two public audio-mesh datasets demonstrate that our method outperforms state-of-the-art techniques both quantitatively and qualitatively in terms of mesh reconstruction accuracy and temporal alignment. In addition, we conducted a user perception study with 20 volunteers to further assess the effectiveness of our approach.

  PublisherDOI

• Level-Set Nonlinear Topology Optimization for Large-Deformation Compliant Mechanisms With Hyperelastic Materials

  2025-08-17

  article

  Abstract The level set method has been widely applied in topology optimization of mechanical structures, primarily for linear materials, but its application to nonlinear hyperelastic materials, particularly for compliant mechanisms, remains largely unexplored. This paper addresses this gap by developing a comprehensive level set-based topology optimization framework specifically for designing compliant mechanisms using neo-Hookean hyperelastic materials. A key advantage of hyperelastic materials is their ability to undergo large, reversible deformations, making them well-suited for soft robotics and biomedical applications. However, existing nonlinear topology optimization studies using the level set method mainly focus on stiffness optimization and often rely on linear results as preliminary approximations. Our framework rigorously derives the shape sensitivity analysis using the adjoint method, including crucial higher-order displacement gradient terms often neglected in simplified approaches. By retaining these terms, we achieve more accurate boundary evolution during optimization, leading to improved convergence behavior and more effective structural designs. The proposed approach is first validated with a mean compliance problem as a benchmark, demonstrating its ability to generate optimized structural configurations while addressing the nonlinear behavior of hyperelastic materials. Subsequently, we extend the method to design a displacement inverter compliant mechanism that fully exploits the advantages of hyperelastic materials in achieving controlled large deformations. The resulting designs feature smooth boundaries and clear structural features that effectively leverage the material’s nonlinear properties. This work provides a robust foundation for designing advanced compliant mechanisms with large deformation capabilities, extending the reach of topology optimization into new application domains where traditional linear approaches are insufficient. The developed methodology is expected to provide a timely solution to computational design for soft robotics, flexible mechanisms, and other emerging technologies that benefit from hyperelastic material properties.

  PublisherDOI

Recent grants

• SGER: Discrete Volumetric Curvature Flow for Graphics Applications

  NSF · $79k · 2008–2009

• Magnetically activated structures for minimally invasive endovascular therapy

  NIH · $627k · 2021–2026

• Collaborative Research: ATD: Algorithmic Aspects of Geometry for Using LIDAR and Wireless Sensor Networks for Combating Chemical Terror Attacks

  NSF · $349k · 2012–2015

• IIS: III: Small: Conformal Geometry for Computer Vision

  NSF · $100k · 2009–2011

• MSPA-MCS: Discrete Curvature Flows on Graphics and Visualization

  NSF · $180k · 2006–2009

Frequent coauthors

• Na Lei

  144 shared

• Wei Zeng

  Hong Kong University of Science and Technology

  109 shared

• Shikui Chen

  Stony Brook University

  103 shared

• Shing‐Tung Yau

  Beijing Institute of Mathematical Sciences and Applications

  86 shared

• Yalin Wang

  72 shared

• Cheng Zhong

  Shenzhen Institutes of Advanced Technology

  64 shared

• J Kim

  64 shared

• Y Kim

  64 shared

Education

• PhD, Computer Science Department

  Harvard University

  2002

• Master, Computer Science Department

  Harvard University

  1996

• BS, Computer Science

  Tsinghua University

  1994

Awards & honors

• Morningside Gold Medal of Applied Mathematics, International…
• Research Excellence Award, Computer Science department, Ston…
• Gaheon Award: The Best Paper in International Journal of CAD…
• Best Paper Award: The 10th International Conference on Compu…
• National Science Foundation (NSF) Faculty Early Career Award…