Xiaosong Wang王潚崧
Xiaosong Wang

Xiaosong Wang

王潚崧 · Ph.D.

Research Scientist, AI for Science (AI4S)

Shanghai Artificial Intelligence Laboratory · 上海人工智能实验室

To advance clinical diagnosis using cutting-edge machine learning techniques.

15,807citations
36h-index
71i10-index
65papers
20patents
Citation statistics from Google Scholar (Nov. 2026).

Brief Bio

I am a Research Scientist in AI for Science at Shanghai Artificial Intelligence Laboratory, working on large-scale medical image/text pre-training and its clinical applications, as well as the Scientific Discovery Platform and Automated Laboratory. Previously, I briefly worked as a Senior Staff Algorithm Engineer at Alibaba DAMO Academy. Before moving back to China, I was a Senior Applied Research Scientist at NVIDIA (2018–2021) and a Visiting Research Fellow in Dr. Ronald Summers's CAD lab at the NIH Clinical Center (2015–2018). From 2011 to 2015 I worked at Shanghai United Imaging Healthcare (UIH), first as CAD Department Manager / Algorithm Engineer and then as Product Manager of multi-modality post-processing workstations. I received my Ph.D. in Computer Science under the supervision of Prof. Majid Mirmehdi at the University of Bristol, UK, in 2011.

My research interests lie in computer vision, vision and language, machine learning and deep learning, and their applications in medical imaging — from large-scale medical image pre-training and foundation models to weakly-supervised learning, lesion detection and clinical datasets.

Recruiting: I am looking for motivated interns, researchers and engineers at all levels. Email me if you are interested.


Research Interests

AI for Science Medical Foundation Models Medical Vision-Language Computer Vision Deep Learning Weakly-Supervised Learning Federated Learning Datasets & Benchmarks Clinical Translation

Education

  • 2006 – 2011
    Ph.D. in Computer Vision
    University of Bristol, UK · Advisor: Prof. Majid Mirmehdi
  • 2005 – 2006
    M.Sc. in Computer Science (Distinction)
    University of Bristol, UK
  • 2001 – 2005
    B.Sc. in Computational Mathematics (Distinction)
    Harbin Institute of Technology, China

Employment

  • Mar. 2022 – present
    Research Scientist, AI4S
    Shanghai Artificial Intelligence Laboratory
    Large-scale medical image/text pre-training and clinical applications; Scientific Discovery Platform and Automated Laboratory.
  • Sep. 2021 – Mar. 2022
    Senior Staff Algorithm Engineer
    Alibaba DAMO Academy
    Led R&D for medical AI solutions and platforms.
  • Aug. 2018 – Aug. 2021
    Senior Applied Research Scientist
    NVIDIA Corporation, Bethesda, USA
    Deep learning solutions for medical image analysis; medical vision & text.
  • Jul. 2015 – Jul. 2018
    Visiting Research Fellow
    CAD Lab, NIH Clinical Center, Bethesda, USA
    Large-scale medical datasets (DeepLesion, ChestX-ray) and deep learning CAD.
  • Dec. 2013 – Jun. 2015
    Product Manager, Post-processing Workstations
    Shanghai United Imaging Healthcare (UIH)
  • Sep. 2011 – Nov. 2013
    CAD Dept. Manager / Tech Lead / Algorithm Engineer
    Shanghai United Imaging Healthcare (UIH)

News

Selected updates — positions, awards, datasets and recent papers.

Research

Seven directions: from AI for Science and medical foundation models to clinical translation, datasets and federated learning.

AI4S

AI4S (AI for Science)

Scientific discovery platform and automated laboratory: LLM agents that reason over wet-lab experiments — bio-experimental protocol generation with structured reward mechanisms, and large-scale LLM routing/aggregation for science workloads.

Representative papers:
  • H. Sun et al., Unleashing Scientific Reasoning for Bio-experimental Protocol Generation via Structured Component-based Reward Mechanism, ICLR, 2026.
  • S. Tang et al., Beyond Gemini-3-Pro: Revisiting LLM Routing and Aggregation at Scale, ICML, 2026.
Medical imaging

Medical Imaging

Core medical image analysis: segmentation (2D/3D, multi-modal, multi-task), detection, weakly-supervised localization, domain generalization and AutoML — from classic CAD (lymph nodes, prostate, lung nodules) to foundation-model-era adaptations.

Representative papers:
  • Z. Tang, X. Wang, Light Yet Effective Adaptations for Multi-modal Multi-task 3D Medical Image Segmentation, ICASSP, 2026.
  • Y. Gao et al., ACDA: Anatomically constrained distribution alignment for robust medical image segmentation, Knowledge-Based Systems, 2026.
  • Y. Gao et al., A Composite Alignment-Aware Framework for Myocardial Lesion Segmentation in Cardiac MRI, MICCAI, 2025.
  • X. Wang et al., Interactive 3D Segmentation Editing and Refinement via Gated GNNs, GLMI-MICCAI, 2019. Best Paper Award
Medical foundation model

Medical Foundation Model

Large-scale medical image/text pre-training and multimodal foundation models: self-supervised pre-training, text-guided adaptation, pathology instruction learning, and medical reasoning — inference-time compute, RL-driven token-sparse reasoning, and mentor-intern collaborative search.

Representative papers:
  • K. Chen et al., Cost-effective instruction learning for pathology vision and language analysis, Nature Computational Science, 2025.
  • H. Sun et al., Chiron-o1: Igniting Multimodal LLMs towards Generalizable Medical Reasoning, NeurIPS, 2025.
  • S. Rui et al., Optimizing Inference-Time Compute for Medical Reasoning via Uncertainty Quantification, ICML, 2026.
  • Z. Luo et al., InvCoSS: Inversion-driven Continual Self-supervised Learning, IEEE CVPR, 2026.
  • X. Wang*, Y. Peng* et al., TieNet, IEEE CVPR, 2018. arXiv
Clinical application

Clinical Application

Translating models into clinical practice: cardiovascular risk prediction from incomplete clinical data, interpretable chest radiograph analysis, focal liver lesion diagnosis on multiparametric MRI, diabetic retinopathy detection, brain disorder diagnosis and multimodal survival analysis.

Representative papers:
  • S. Rui et al., Predicting major adverse cardiovascular events from incomplete clinical data, npj Digital Medicine, 2026.
  • L. Xu et al., A unified multi-task framework enables interpretable chest radiograph analysis, Cell Med, 2026.
  • Z. Shen et al., An Explainable Deep Learning Model for Focal Liver Lesion Diagnosis Using Multiparametric MRI, Radiology: Artificial Intelligence.
  • Z. Tang et al., Reasoning-Enhanced VLM for Interpretable Diabetic Retinopathy Detection, MICCAI-OMIA, 2025. Best Oral Paper Award
  • S. Rui et al., CardioCoT: Hierarchical Reasoning for Multimodal Survival Analysis, ICASSP, 2026.
Dataset and benchmark

Dataset and Benchmark

Curating large-scale public datasets and benchmarks: ChestX-ray14, DeepLesion, a multimodal retinal dataset for diabetic retinopathy, a multi-phase liver CT dataset, and a real-world benchmark for foundation model adaptation.

Representative papers:
  • Z. Tang et al., A multimodal retinal image dataset for diabetic retinopathy detection using foundation models, Scientific Data, 2026.
  • X. Wu et al., A Multi-phase CT Dataset for Automated Differential Diagnosis of Liver Tumors, Scientific Data, 2025.
  • D. Wang*, X. Wang* et al., A Real-world Dataset and Benchmark for Foundation Model Adaptation, Scientific Data, 2023.
  • K. Yan, X. Wang et al., DeepLesion, Journal of Medical Imaging, 2018. Data
  • X. Wang et al., ChestX-ray8, IEEE CVPR (Spotlight), 2017. arXiv · Data
Federated learning

Federated Learning

Privacy-preserving collaborative learning across institutions: modality-agnostic heterogeneous federated learning, reinforcement-learning hyperparameter control in asynchronous federated learning, federated prostate and COVID-19 segmentation, and straggler-aware optimization.

Representative papers:
  • W. Zhang et al., Modality-Agnostic Federated Learning with Adaptive Updates for Heterogeneous Medical Image Tasks, IEEE TMI, 2026.
  • S. Gu et al., Dynamic Hyperparameter Adjustment via RL in Asynchronous Federated Learning, MICCAI-DeCaF, 2025.
  • D. Yang et al., Federated semi-supervised learning for COVID region segmentation, Medical Image Analysis, 2021.
  • H. R. Roth et al., Federated Whole Prostate Segmentation in MRI, MICCAI, 2021.
Other algorithmic research

Other Algorithmic Research

Algorithm work beyond the medical domain: probabilistic archive film restoration (HMM defect detection, MRF false-alarm elimination, random-walk restoration), robust label-quality learning, and generative radiogenomic mapping.

Representative papers:
  • X. Wang, M. Mirmehdi, Archive Film Defect Detection and Removal, IEEE TIP, 2012.
  • X. Wang, M. Mirmehdi, HMM based Archive Film Defect Detection, BMVC, 2009. Best Industrial Paper Award
  • X. Wang et al., Learning Quality Labels for Robust Image Classification, WACV, 2024.
  • Z. Xu et al., Correlation via Synthesis: End-to-end Image Generation and Radiogenomic Learning, MIDL, 2020.

Publications

Full list — journal papers, conference papers, book chapters, preprints, abstracts and patents.

0 results

Awards

Honors and recognitions.

2025
Best Oral Paper Award
OMIA-MICCAI 2025
2023–25
Elsevier Most Cited Chinese Researchers (Computer Science)
2019
Best Paper Award
GLMI-MICCAI 2019
2017
NIH Clinical Center Director’s Award
NIH Clinical Center
2016
RSNA Roentgen Fellow Research Award 2016
Radiological Society of North America
2016
NIH Fellows Award for Research Excellence (FARE)
NIH
2010
ECCV Student Travel Award
ECCV 2010
2009
Best Industrial Paper Prize
BMVC 2009
2006–09
Great Western Research Ph.D. Scholarship
United Kingdom
2006
M.Sc. thesis with Distinction (top 5%)
University of Bristol
2005
B.Sc. thesis with Distinction
Harbin Institute of Technology

Professional Service

Editorial

  • Associate Editor — IEEE Journal of Biomedical and Health Informatics
  • Associate Editor — Frontiers in Nuclear Medicine
  • Guest Editor — Medical Image Analysis

Conference Reviewing

  • Area Chair / Senior PC — NeurIPS, IJCAI, AAAI, MICCAI, ICLR, MIDL
  • Reviewer — CVPR, ICCV, ECCV, MICCAI, NeurIPS, ICML, AAAI, IJCAI, ICLR

Journal Reviewing

Nature Machine Intelligence · Nature Communications · Lancet Digital Health · IEEE Transactions on Medical Imaging · Pattern Recognition · Medical Image Analysis · IEEE Transactions on Biomedical Engineering · IEEE Journal of Biomedical and Health Informatics · Journal of Biomedical Informatics · IET Computer Vision · Neurocomputing · Knowledge-Based Systems · Machine Vision and Applications · Pattern Analysis and Applications

Datasets & Software

Open datasets and benchmarks built and released by our group.

ChestX-ray14

Hospital-scale chest X-ray database: 108,948 frontal-view X-rays of 32,717 patients with text-mined labels for 14 common thorax diseases. CVPR 2017 (Spotlight).

Dataset Paper

DeepLesion

Large-scale clinical lesion dataset: 33,688 bookmarked radiology images with lesion annotations, mined from hospital PACS. Journal of Medical Imaging, 2018.

Dataset Paper

Multimodal Retinal DR Dataset

Multimodal retinal imaging dataset for diabetic retinopathy detection using foundation models (Scientific Data, 2026).

Multi-phase Liver CT Dataset

Multi-phase CT dataset for automated differential diagnosis of liver tumors (Scientific Data, 2025).

More

Mentorship, invited talks and media coverage.

Mentorship

Yixiao ZhangIntern · Johns Hopkins University (2019 @NVIDIA)
Liyue ShenIntern · Stanford University (2019 @NVIDIA)
Puyang WangIntern · Johns Hopkins University (2020 @NVIDIA)
Riddhish BhalodiaIntern · University of Utah (2020 @NVIDIA)
Shiyi DuIntern · Sichuan University (2022–23 @ShAILab)
Linghao QuIntern · Fudan University (2023–24 @ShAILab)
Haoyang SuIntern · Fudan University (2024–26 @ShAILab)
Haoran SunIntern · Fudan University (2025– @ShAILab)
Yifan GaoMentee · Shanghai Innovation Institute (2024–26)
Jie YangMentee · Shanghai Innovation Institute (2024–26)
Yujie ZhangMentee · Shanghai Innovation Institute (2025–26)
Wenjie LiMentee · Shanghai Innovation Institute (2025–26)
Zhenyu TangPhD student · Shanghai Jiao Tong University (2022– )
Shaohao RuiPhD student · Shanghai Jiao Tong University (2023– )
Kaitao ChenPhD student · Fudan University (2024– )
Shuang GuPhD student · Shanghai Jiao Tong University (2024– )
Xiangyu SunPhD student · Shanghai Jiao Tong University (2026– )

Invited Talks

  • Jul. 2018 — Mine Deeper and Learn Wider: a Perspective on Distilling Radiological Reports for Chest X-ray Analysis. Medical Computer Vision and Health Informatics Workshop, CVPR 2018.
  • Jul. 2017 — Big Data, Weak Label and True Clinical Impacts for Radiology Imaging Diagnosis. Medical Computer Vision and Health Informatics Workshop, CVPR 2017.
  • Jun. 2017 — Weakly supervised Classification and Localization of Common Chest Disease in X-ray using Deep Learning. NCBI, National Library of Medicine.
  • Feb. 2017 — Unsupervised Image Categorization in a Large Scale Radiology Image Database. Computer Science Department, Johns Hopkins University.
  • Mar. 2016 — Towards large-scale radiology image database auto-annotation: a Deep Pseudo-task Learning approach. NIH Pi Day, PiCo Talks 2016.

Media Coverage

  • Oct. 2017 — NIH releases massive database of chest x-rays. Health Data Management.
  • Oct. 2017 — NIH Releases Chest X-Rays + Data. Federal Telemedicine News.
  • Sep. 2017 — How 100K Chest X-Ray Images Could Improve AI and Patient Health. Healthcare Analytics News.
  • Sep. 2017 — NIH releases massive database of chest x-rays. AuntMinnie.
  • Sep. 2017 — NIH to provide one of largest chest x-ray datasets for research. United Press International.
  • Sep. 2017 — NIH Clinical Center provides one of the largest publicly available chest x-ray datasets. NIH news release.
  • Sep. 2017 — NIH Releases 100,000 Chest X-Rays. HealthLeaders Media.
  • Sep. 2017 — NIH Clinical Center issues 100K x-ray images for machine learning. Radiology Business.
  • Oct. 2016 — Deep learning can find, label images in PACS. AuntMinnie.
  • Dec. 2009 — Summary of Prizes Presented at BMVC 2009. British Machine Vision Association News.