Yiding Gu / 顾一丁

个人学术主页 Academic Profile

个人简介 Personal CV

2027 秋季 博士申请 Ph.D. Application for Fall 2027

2019–2024 本科 · 临床医学（八年制）
中国医科大学 Bachelor's Degree · Clinical Medicine (Eight-Year Program)
China Medical University
2024–2027 硕士（专业学位）· 泌尿外科
中国医科大学 Master's Degree (Professional) · Urology
China Medical University

专注于临床医学与泌尿外科领域的学习与实践。在这里记录我的发表研究、进行中项目，以及工作经历。 Focused on clinical medicine and urology. Here I document my published research, ongoing projects, and work experience.

顾一丁证件照
Yiding Gu 外科学硕士研究生 Master's Student in Surgery
顾一丁微信二维码 扫码添加微信 Scan to connect on WeChat

已发表研究 Published Research

Academic Radiology, 2025

MCANet：用于 pSA-AKI 早期预测的多模态交叉注意力网络 MCANet: Multimodal Cross-Attention Network for pSA-AKI prediction

论文：Integrating Multi-Modal Imaging Features for Early Prediction of Acute Kidney Injury in Pneumonia Sepsis: A Multicenter Retrospective Study Paper: Integrating Multi-Modal Imaging Features for Early Prediction of Acute Kidney Injury in Pneumonia Sepsis: A Multicenter Retrospective Study 论文链接 Paper link

  • 胸部 CT Chest CT
  • 多模态融合 Multimodal fusion
  • 模型可解释性 Interpretability analysis
  • 设计并实现多模态融合分类模型，整合肺部、心外膜脂肪组织（EAT）和 T4 水平皮下脂肪组织（T4-SAT）的区域特异性特征。 Built the MCA-Net multimodal fusion classification model, integrating region-specific features from the lung, epicardial adipose tissue (EAT), and T4-level subcutaneous adipose tissue (T4-SAT).
  • 第一阶段使用 ResNet-18 从各区域 2D CT 切片中提取高维影像特征，随后进行跨模态相似性建模和注意力重加权。 Used ResNet-18 in the first stage to extract high-dimensional features from 2D CT slices of each region, followed by MSFAN-based cross-modal similarity computation and attention reweighting.
  • 在独立中心外部测试集上，Lung + T4-SAT + EAT 三模态模型取得 0.981 的准确率和 0.99 的 AUC。 On the external test set from an independent center, the Lung + T4-SAT + EAT three-modality model achieved an accuracy of 0.981 and an AUC of 0.99.

原文图片 Original Figures

Academic Radiology Figure 2
图 2. MCA-Net 架构：基于 ResNet-18 的区域特征提取、MSFAN 跨模态注意力融合和 ResNet-101 分类。 Fig. 2. MCA-Net architecture: ResNet-18-based regional feature extraction, MSFAN cross-modal attention fusion, and ResNet-101 classification.
Academic Radiology Figure 4
图 4. 深度特征与临床变量的相关性分析，以及肺部、EAT 和 T4-SAT 区域的 Grad-CAM 注意力图。 Fig. 4. Correlations between deep features and clinical variables, with Grad-CAM attention maps for the lung, EAT, and T4-SAT regions.

学术会议 Conferences

🏆 RSNA 2026 · 北美放射学会年会 🏆 RSNA 2026 · Radiological Society of North America Poster Discussion（口头壁报讨论） Poster Discussion

Foundation Model-Based Multimodal AI for Prostate Cancer Risk Prediction

📍 芝加哥，伊利诺伊州 Chicago, IL 📅 Dec 3, 2026 ⏱ 12:15–12:45 PM30 分钟讨论30 min discussion

作者： Authors: Yiding Gu, Tianming Du, Langjun Li, Xiaofeng Qu

  • 入选为 Top Submission，获 30 分钟 Poster Discussion。 Selected as a Top Submission from 12,000+ abstracts for a 30-minute Poster Discussion presentation.
  • 基于多中心前列腺 MRI 数据，构建基础模型预训练（MAE + BYOL）+ ClinicalBERT 多模态融合框架。 Built a foundation model (MAE + BYOL) + ClinicalBERT multimodal fusion framework on ~6,000 multi-center prostate MRI cases.
  • BPH vs PCa 鉴别 AUC 达 0.854，优于纯影像模型（0.848）和纯临床模型（0.81），展现多模态互补优势。 BPH vs PCa discrimination AUC 0.854, outperforming imaging-only (0.848) and clinical-only (0.81) models, demonstrating multimodal complementarity.
  • 集成 LangChain AI Agent 实现指南指导的临床决策支持，提供可审计的诊疗建议。 Integrated LangChain AI Agent for guideline-grounded clinical decision support with auditable recommendations.
  • 基础模型 Foundation Model
  • 多模态融合 Multimodal Fusion
  • 前列腺 MRI Prostate MRI
  • 临床决策支持 Clinical Decision Support
  • AI Agent AI Agent

在研课题 Ongoing Work

已提交 · 审稿中Submitted · Under review

ProstateMind for selective evidence-grounded prostate disease decision support ProstateMind for selective evidence-grounded prostate disease decision support

摘要
前列腺疾病人工智能系统的临床价值不仅取决于预测准确性，还取决于其处理证据缺失或相互矛盾的能力。我们开发了 ProstateMind，一个整合自监督 MRI 表征、任务特异性风险模型、确定性路径分流和指南证据驱动检索增强生成的模块化工作流。在来自六个数据源的 4,880 次检查中，模型在内部测试集检出前列腺癌和临床显著性前列腺癌的曲线下面积分别为 0.879 和 0.924，在外部队列检出临床显著性前列腺癌的曲线下面积为 0.833。混合检索联合重排序器在 100 个问题上的 Hit@10 和 Recall@10 分别为 0.960 和 0.925。三名专家评价了 50 个生成答案；150 次评分事件中有 8 次被标记为存在幻觉，未发现不安全建议。在前瞻性评价的 193 例患者中，系统将 22 例（11.4%）分配至疾病特异性路径，将 171 例（88.6%）保留在综合评估路径。上述结果支持将概率风险估计与临床路径承诺相分离，并通过可追溯证据和临床医生监督约束系统输出。
关键词：前列腺癌；磁共振成像；自监督学习；临床决策支持；检索增强生成；不确定性分流。 Abstract
Clinical value of artificial intelligence for prostate disease depends not only on predictive accuracy but also on handling missing or conflicting evidence. We developed ProstateMind, a modular workflow integrating self-supervised magnetic resonance imaging representations, task-specific risk models, deterministic routing and guideline-grounded retrieval-augmented generation. Across 4,880 examinations from six data sources, internal areas under the curve for prostate cancer and clinically significant prostate cancer detection were 0.879 and 0.924, respectively; the external area under the curve for clinically significant disease was 0.833. Hybrid retrieval with reranking achieved Hit@10 of 0.960 and Recall@10 of 0.925. Three experts evaluated answers to 50 questions; 8 of 150 ratings flagged potential hallucination and none flagged unsafe advice. Among 193 prospectively evaluated patients, 22 (11.4%) entered disease-specific pathways and 171 (88.6%) remained under integrated assessment. These findings support separating probabilistic risk estimation from pathway commitment while constraining outputs through traceable evidence and clinician oversight.
Keywords: prostate cancer; magnetic resonance imaging; self-supervised learning; clinical decision support; retrieval-augmented generation; selective routing.

  • 前列腺癌 Prostate cancer
  • 自监督学习 Self-supervised learning
  • 检索增强生成 Retrieval-augmented generation
  • 临床决策支持 Clinical decision support
  • 磁共振成像 Magnetic resonance imaging
  • 不确定性分流 Selective routing
ProstateMind Figure 1
ProstateMind Figure 2
ProstateMind Figure 3
ProstateMind Figure 4
ProstateMind Figure 5
ProstateMind Figure 6
1 / 6 / 6

脂肪先验引导的 CT-to-PET 跨模态合成，用于临床可及的代谢评估 Adipose-Prior-Guided CT-to-PET Cross-Modality Synthesis for Clinically Accessible Metabolic Assessment

GitHub： GitHub: github.com/llj0621/Adipose_Prior_CT2PET

  • CT-to-PET 合成 CT-to-PET synthesis
  • 3D 条件 GAN 3D conditional GAN
  • 脂肪先验 Adipose priors
  • 代谢风险评估 Metabolic risk analysis
Adipose-Prior-Guided CT-to-PET framework

资源与合作Resources & Collaboration

专用 AI 算力基础设施Dedicated AI Computing Infrastructure

支持临床 AI 研究与医学转化的高性能计算平台High-performance computing for clinical AI research and medical translation

我的研究由专用 AI 算力基础设施支持，目前包括三台高性能服务器，每台配备七张 NVIDIA GeForce RTX 5090 GPU（每张 32 GB 显存）。该平台由我与合作伙伴共同搭建，我参与了从机架安装、硬件组装到系统部署的完整流程。这一经历使我积累了高性能 AI 计算系统部署、运行和维护的实践经验。My research is supported by dedicated AI computing infrastructure comprising three high-performance servers, each equipped with seven NVIDIA GeForce RTX 5090 GPUs with 32 GB of memory per GPU. I co-built the platform with partners and participated throughout the process, from rack installation and hardware assembly to system deployment, gaining practical experience in deploying, operating, and maintaining high-performance AI computing systems.

该算力基础设施归属于下表所列的赫科文化科技（沈阳）有限公司，我作为创始成员之一参与建设。结合与合作医院及青年研究团队的长期协作，该平台支持从临床数据获取、AI 算法开发、模型验证到医学转化研究的完整工作流。The infrastructure belongs to Heke Culture Technology (Shenyang) Co., Ltd., listed below, and I contributed to its development as a founding member. Together with long-term collaborations involving partner hospitals and early-career research teams, the platform supports an end-to-end workflow spanning clinical data acquisition, AI algorithm development, model validation, and translational medical research.

AI GPU computing infrastructure
专用 GPU 算力基础设施实景Dedicated GPU computing infrastructure

合作机构Collaborating Institutions

合作机构Institution机构类型Institution type合作内容Collaboration
沈阳材料科学国家研究中心，中国科学院金属研究所Shenyang National Laboratory for Materials Science, Institute of Metal Research, Chinese Academy of Sciences材料科学研究机构Materials science research institute围绕器件-AI交叉方向开展光电计算、器件表征与算法应用合作。Collaboration in optoelectronic computing, device characterization, and algorithm applications at the device–AI interface.
中国医科大学附属盛京医院（辽宁，中国）Shengjing Hospital of China Medical University (Liaoning, China)大学附属医院University-affiliated hospital临床影像数据、多中心队列建设合作。Clinical imaging data and multicenter cohort development.
大连医科大学附属第二医院（辽宁，中国）Second Affiliated Hospital of Dalian Medical University (Liaoning, China)大学附属医院University-affiliated hospital临床影像数据、多中心队列建设合作。Clinical imaging data and multicenter cohort development.
兴安盟人民医院（内蒙古，中国）Hinggan League People's Hospital (Inner Mongolia, China)区域医院Regional hospital区域临床队列和外部验证场景建设合作。Regional clinical cohort development and external validation settings.
海城市正骨医院（辽宁，中国）Haicheng Orthopedic Hospital (Liaoning, China)区域专科医院Regional specialty hospital面向专科临床问题的合作研究和临床转化中心。Collaborative research and a clinical translation center focused on specialty clinical questions.
本溪市中心医院（辽宁，中国）Benxi Central Hospital (Liaoning, China)区域三甲医院Regional tertiary hospital泌尿外科多中心队列建设合作。Multicenter urology cohort development.
丹东市中心医院（辽宁，中国）Dandong Central Hospital (Liaoning, China)区域三甲医院Regional tertiary hospital泌尿外科多中心队列建设合作。Multicenter urology cohort development.
东北大学（中国）Northeastern University (China)研究型大学Research university跨学科AI方法、算法验证和工程实现合作。Interdisciplinary AI methods, algorithm validation, and engineering implementation.
吕贝克大学（德国）University of Lübeck (Germany)研究型大学Research university跨学科AI方法、算法验证和工程实现合作。Interdisciplinary AI methods, algorithm validation, and engineering implementation.
赫科文化科技（沈阳）有限公司Heke Culture Technology (Shenyang) Co., Ltd.初创公司Startup company算力基础设施、AI工程部署和医学AI转化支持。Computing infrastructure, AI engineering deployment, and medical AI translation support.

手机端可左右滑动查看完整表格Swipe horizontally on mobile to view the full table

荣誉证书 Honors & Certificates

🩺
医师资格证 Physician Qualification Certificate 专业资质 Professional Qualification
🎓
研究生一等奖学金 First-Class Graduate Scholarship 学业荣誉 Academic Honor
🌐
大学英语六级 CET-6 英语能力证书 English Proficiency
📖
大学英语四级 CET-4 英语能力证书 English Proficiency

保持联系 Get in touch

正在申请 2027 年秋季博士，期待与您交流。 Seeking Ph.D. positions for Fall 2027. I welcome your contact.

所在地： Location: 辽宁 · 沈阳 Shenyang, Liaoning 电话： Phone: 15102446857 邮箱： Email: guyiding1010@163.com