Li Chen Curriculum Vitae


Last update: Oct. 2026


RESEARCH INTERESTS

Medical Image Analysis, Machine Learning, Computer Vision


ACADEMIC QUALIFICATIONS

Sep. 2016 – Aug. 2021

Ph.D. in Electrical and Computer Engineering, University of Washington
Dissertation: Feature extraction and quantification to explore human vasculature. (PPT) (PDF)
Research assistant in Vascular Imaging Lab. Co-advised by Prof. Chun Yuan and Prof. Jenq-Neng Hwang
Overall   GPA:  3.89 /4.0

Sep. 2012 – June 2016

B.S. in Electrical Engineering, Fudan University
Overall   GPA:  3.6 /4.0      Ranking:  9%

Jan. 2015 – May. 2015

Exchange Student in National University of Singapore
Overall   GPA:  5.0 /5.0


WORK EXPERIENCES

Oct. 2021 – Now

Research Scientist, Philips North America
Ultrasound AI Group, Philips Research / Philips Ultrasound, Cambridge, MA. Managed by Dr. Alvin Chen.

May – Aug. 2021

Research Intern, Genentech, Inc.
Early Clinical Development Informatics (ECDi) group. Advised by Dr. Reza Negahdar.

June – Sep. 2020

Research Intern, United Imaging Intelligence America
Medical AI group. Advised by Dr. Shanhui Sun and Terrence Chen.


CONTRIBUTIONS

1. Clinical translation: Translating AI into clinically useful and deployable medical technologies

I develop medical AI systems with an emphasis on translating algorithms into practical clinical tools, addressing not only model performance but also data limitations, annotation cost, workflow integration, and clinical validation.

At Philips Research, I have contributed to the full development cycle of AI for point-of-care lung ultrasound, from clinical problem formulation and algorithm development through function-proven milestones, clinical evaluation, and commercial deployment on the Lumify ultrasound platform. My work includes weakly and semi-supervised learning to reduce dependence on expensive frame-level annotations, contrastive self-supervised learning to exploit large collections of unlabeled ultrasound videos, and graph-based models for reasoning from detections to videos and patients. These approaches enabled lung ultrasound AI for consolidation and pleural effusion detection while substantially reducing annotation requirements and improving diagnostic performance. I also contributed to multi-reader multi-case clinical validation study demonstrating improved reader performance and reduced interpretation time with AI assistance.

Earlier work applied computer vision and data-mining techniques to traditional Chinese medicine, including automated tongue segmentation and computer-aided tongue diagnosis, illustrating my longstanding interest in translating computational methods into clinically relevant tools.

In vascular imaging, I developed LATTE, a fully automated MRI-based workflow for screening carotid atherosclerotic lesions, designed to reduce the burden of manual image review and assist clinicians in identifying suspicious regions.

2. Technical innovation: Knowledge-guided AI for challenging medical imaging problems

A central theme of my research is incorporating medical knowledge and structural constraints into AI models rather than treating medical images as generic visual data. I develop methods that explicitly exploit anatomy, geometry, shape, topology, and spatiotemporal relationships to solve difficult medical-image analysis problems.

For vascular imaging, I developed graph-based methods for anatomical labeling of intracranial arteries, explicitly modeling hierarchical anatomical relationships to achieve accurate artery identification. I subsequently developed vessel-wall segmentation methods that exploit the approximately circular geometry of vessels through polar-coordinate representations, and developed artery tracing methods that incorporate active-contour shape constraints into deep learning frameworks.

This knowledge-guided perspective has continued in my ultrasound research, where graph neural networks model relationships among spatially and temporally connected detections and anatomical lung zones, allowing clinically meaningful aggregation of sparse findings distributed across ultrasound videos and examinations.

More broadly, my work demonstrates how domain knowledge can be converted into computational structure to improve robustness, data efficiency, interpretability, and performance in medical AI.

3. Scientific discovery: AI-enabled quantitative imaging for discovering new medical knowledge

Beyond developing predictive models, I build automated image-analysis tools that make previously difficult or impractical clinical research feasible at scale.

During my Ph.D., I developed iCafe (intraCranial Artery Feature Extraction), a comprehensive software platform for tracing, modeling, labeling, and quantifying the cerebral arterial network. By converting complex vascular images into structured quantitative representations, iCafe enabled reproducible analysis of cerebral vascular morphology across large cohorts and has been used by 34+ sites in 9+ countries. This infrastructure supported studies revealing associations between vascular morphology and aging, cognitive impairment, and vascular risk.

I similarly developed FRAPPE, a fully automated pipeline for popliteal artery and vessel-wall analysis, and applied it to approximately 3.5 million images from the Osteoarthritis Initiative—an analysis scale that would otherwise require decades of manual expert review. This enabled population-scale investigation of vascular characteristics and revealed differences in vessel-wall measurements associated with cardiovascular risk.

Across these projects, I use AI and quantitative image analysis not only to automate existing clinical tasks, but also as scientific instruments for extracting measurable biological information and uncovering previously difficult-to-observe patterns in human disease.


RESEARCH EXPERIENCES

Philips Research North America

Oct. 2021 ~ Now

Research Scientist. Ultrasound AI group.

Cambridge, MA

  • Lung ultrasound AI for Philips Lumify (point-of-care imaging device)

    • Spearheaded full-cycle development of commercial consolidation and pleural effusion detection features, from concept selection through function-proven milestones and clinical validation toward commercial deployment.
    • Built a teacher-student mutual-learning framework for frame-labeled, video-labeled, and unlabeled data, reducing estimated annotation effort for 15,000 videos from 15,000 hours to about 250 hours.
    • Contrastive pretraining on 27,063 unlabeled videos from 528 patients raised test AUC from 0.65 to 0.88 using 5% of labeled data (IEEE ISBI 2023).
    • In a multi-reader, multi-case study, AI assistance met non-inferiority and superiority endpoints, improved reader-averaged AUC by 0.101/0.046 for consolidation/effusion, and reduced reading time by 16.6% (12.5 seconds).

  • Graph neural networks for ultrasound reasoning at multiple levels

    • Detection: A graph-based pseudo-label selection method achieved mAP@0.5 of 0.586/0.439 on breast/lung datasets with limited box labels (IEEE Transactions on Medical Imaging, conditionally accepted).
    • Video: GADA aggregates frame-level detections and spatiotemporal relationships, reaching AUCs of 0.943/0.960 for consolidation/effusion across six sites and 9,900+ lung ultrasound videos (ICCV 2025 Workshop).
    • Patient: ZAGNet models relationships among lung-zone findings, improving patient-level AUC from 0.677/0.815 with pooling to 0.803/0.893 for consolidation/effusion (IEEE IUS 2026).
    • Filed a series of invention disclosures and patents (12 with major contributions), strengthening Philips' intellectual property portfolio.

Professional service: Chair for the Deep Learning section in Philips' intellectual property review group.

Vascular Imaging Lab & Information Processing Lab, University of Washington

Sep. 2016 ~ Aug. 2021

Research Assistant. Co-advised by Prof. Chun Yuan, Department of Radiology and Bioengineering, and Prof. Jenq-Neng Hwang, Department of Electrical and Computer Engineering

Seattle, WA

  • Quantitative intracranial artery modeling and vascular feature extraction

    • Developed an artery tracing and labeling tool (iCafe, C++ software, 60k lines) to model arteries and extract cerebral vascular features.
    • A novel artery refinement algorithm through optimization on Curved Planar Reformation view.
    • A novel artery naming algorithms using Graph Neural Network and hierarchical refinement.
    • Created a database of 1000+ scans of cerebral vasculature models.
    • Tool used by 12+ sites on dozens of medical research researches (aging, dementia, artery revascularization, etc.).
    • Three first-author journal papers (1,2,3) and seven conference publications ranging from technical development, validation and medical applications.
    • Editor's pick by Magnetic Resonance in Medicine.
    • iCafe website: http://icafe.clatfd.cn

  • Vessel wall segmentation using convolutional neural networks

    • Proposed an automated vessel wall segmentation and quantification method. (paper)
    • 3D region of interest identification using object tracking (Yolo V2 detector and tracklet refinement).
    • A novel vessel wall segmentation algorithm in polar coordinate system using deep learning.
    • Segmentation with uncertainty scores, proved to be useful in indicating segmentation performance.
    • Trained on more than one thousand subjects of labeled carotid vessel wall contours using convolutional neural network (CNN). Test set performance better than traditional cartesian based CNN methods (U-Net, Mask-RCNN).

  • Automated popliteal vessel wall segmentation and quantification.

    • Developed a fully-automated and robust vessel wall analysis tool (FRAPPE) to segment and quantify popliteal arteries and vessel walls for vascular research. (paper)
    • Processed on The Osteoarthritis Initiative (OAI) dataset with 3.5 Million popliteal artery images.
    • Transfer learning (from the carotid artery model) and active learning techniques to reduce labeling burden while maintaining accuracy.
    • Dice of 0.79 with human contours, only 1.2% images have major errors.
    • Found significant differences in vessel wall thickness measurements between high and low risk subjects.
    • Multi GPU process within 2 months, the workload of 70 years for an expert human reader.
    • Winner for American Heart Association/Amazon Web Services Prize Competition.
    • More technical details:http://clatfd.cn/a/286

  • Carotid artery atherosclerotic lesion screening using an AI based fully automated workflow based on 3D MRI

    • Proposed a 5-minute automated Magnetic Resonance screening workflow using multiple deep learning models.
    • 3D MERGE as the fast (2 minutes) MR imaging sequence.
    • Image quality assessment using target weighted patches. (joint work with Hongjian Jiang, master student under mentorship)
    • Multi slice multi channel patches for lesion classification.
    • High agreements with an expert radiologist (0.9+ sensitivity/specificity).
    • Assist radiologists by warning potential locations of advance lesions visualized in 3D view.
    • Effective in reducing manual review time for both experts and novice radiologists in clinical reading.
    • Online learning method to reduce labeling labors.
    • Patent filed.

Genentech, Inc.

May ~ Aug. 2021

Research intern. Advised by Dr. Reza Negahdar, Early Clinical Development Informatics (ECDi) group.

South San Francisco, CA

  • Imaging and non-imaging feature fusion

    • Developed a feature fusion model for improving COVID-19 classification from CTA images.
    • Fusion of imaging and non-imaging features improves COVID classification by 6.6%.
    • Grad-CAM to interpret image region importance.

United Imaging Intelligence America

June ~ Sep. 2020

Research intern. Advised by Dr. Shanhui Sun and Terrence Chen, Medical AI group

Boston, MA

  • Landmark tracking for stent enhancement

    • A deep learning workflow for balloon marker detection, tracking and stent enhancement.
    • Developed a marker detector with 95% recall / 50% precision / 44 frames per second.
    • Robust marker tracking using graph search with 0.36mm mean distance with manual labels.
    • Iterative process to separate stent and clutter layer for stent enhancement.

Circuit Theory and Application Lab, Fudan University

Oct. 2013 ~ June 2016

Undergraduate Researcher. Advised by: Dr. Huiliang Shang, Associate Professor, Department of Electronic Engineering

Shanghai, China

  • Capstone research: 'Vascular image registration by circuit simulation'(published in BMC bioinformatics, First Author)

    • A novel algorithm to represent vasculatures as circuits for robust matching.
    • Convert vascular trees into circuits, register using simulated voltages
    • Use Network Structure Index to represent regional graph features
    • Robust registration method tolerates noises and distortions
    • Best undergraduate thesis in Fudan University

  • Research: 'A Novel Automatic Tongue Image Segmentation Algorithm: Color Enhancement Method Based on L*a*b* Color Space'(published in Bioinformatics and Biomedicine (BIBM), 2015 IEEE International Conference on, First Author)

    • Utilize potential advantages of specific characteristics in three color-spaces to find a new method to detect tongue contour in the tongue diagnosis image.
    • In HSV space, using a hue threshold control function to determine a proper threshold value for preliminary separation of tongue region
    • In RGB space, performing color enhancement to obtain a better luminance which is more suitable for tongue segmentation
    • In L*A*B* space, using the luminance sensitivity of L channel to constrain the area of interest
    • Cooperated with Shanghai Traditional Chinese Medicine University and series of local hospitals

  • Research: 'The Application Characteristics of Traditional Chinese Medical Science Treatment on Vertigo Based on Data Mining Apriori Algorithm' (published on IJWMC, First Student Author)

    • To investigate statistical law in Traditional Chinese Medical(TCM) Treatment on vertigo using data mining technique
    • Created a database of TCM treatment prescriptions where 100 cases of prescriptions are selected.
    • Using Apriori algorithm for association rules mining in SPSS to find interrelationships among TCM syndromes, symptoms, and herbal medicine for vertigo, helping research on vertigo using TCM
    • Proposed a topology representation method using circuit simulation for vascular registration.

  • Research: 'An Adaptive Computer-aided Tongue Diagnosis Method using Color-calibration Preprocessing and Multiple Feature Synthesis based on Android' (published on IJWMC, First Student Author)

    • A research on diagnosing disease by taking a picture from tongue based on Android phone.
    • Accuracy and effectiveness corroborated with professional clinicians
    • Used QCGP to solve the problem of white balance
    • Adapted HSV model in Tongue brim pixels searching
    • Adapted inverted pentagonal four-line tongue outline searching and linking algorithm

Adaptive Networks and Control Lab (CAN), Fudan University

Mar. 2014 ~ Mar. 2015

Undergraduate Researcher. Advised by: Prof. Xiang Li, Professor, Department of Electronic Engineering

Shanghai, China

  • Research: 'An improved method of acquaintance immunization strategy in complex network'(published on JTB (IF:2.116), First Author)

    • A research to find an improved method to effectively immune a virus spreading in complex network.
    • A new Index of NSI is presented to value the structure and importance of nodes in network
    • Improved the classical acquaintance immunization strategy using NSI to protect 14.9% more nodes and decrease the spread rate by 27.6% (compared to classical strategy in our simulation)
    • Simulated in various network structures. Best in most random graph compared to other immunization strategies in the perspective of maximum percentage of infected nodes

Key Laboratory of EMW information, Fudan University

Oct. 2013 ~ Nov. 2014

Undergraduate Researcher. Advised by: Prof. Bin Wang, Professor, Department of Electronic Engineering

Shanghai, China

  • FDUROP Project: 'The development of an inaccurate graph isomorphism algorithm and its application in prototype machine' (Exhibited on 2014 Shanghai Industrial Exhibition)

    • Building a prototype machine which uses contour extraction algorithms to determine the location of products in images captured from production line and sort them according to their shape.
    • Funded 6000 RMB by FuDan Undergraduate Research Opportunity Program(FDUROP)
    • Cooperated with Shanghai GO-WELL Electrical Technology CO.
    • Used Matlab and OpenCV for the algorithm and executed on multi-platform (PC, raspberry, DSP)
    • Endured noises and overlaps, high recognition of different products on the same production line

Fudan Physic Teaching Lab, Fudan University

Sep. 2013 ~ Dec. 2013

Undergraduate Researcher. Advised by: Prof. Yongkang Le, Professor, Department of Physics

Shanghai, China

  • Application: 'A remote & online electrical control center using infrared ray based on Arduino'

    • A versatile remote control center which can operate multiple electrical devices using infrared ray.
    • Emit infrared ray instructions under Internet control from remote devices (PC, smartphone, iPad)
    • Can memorize codes of up to 3 infrared control devices
    • A project for the Arduino contest held by Fudan Physic research center

For Whole Research Experience, Link to Project Journey


PUBLICATION

Journal Publications


Conference Publications


Conference/Workshop Oral Presentation


Invited Talks


Preprints


Conference Posters/E-posters


Patents


ACADEMIC SERVICE

Journal Reviewer IEEE Transactions on Medical Imaging, Journal of Biomedical and Health Informatics, Computers in Biology and Medicine, IEEE Access, Artificial Intelligence in Medicine, Medical & Biological Engineering & Computing, The International Journal of Cardiovascular Imaging, Stroke and Vascular Neurology, American Journal of Neuroradiology, Magnetic Resonance Imaging, Electronics and Telecommunications Research Institute (ETRI) Journal, PLOS ONE, International Journal of Modern Physics B.
Conference Reviewer the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), The AICity 2020 CVPR Workshop.
Challenge Organizer Carotid Artery Vessel Wall Segmentation Challenge. (Endorsed by SMRA 2021 and MICCAI 2021)


AWARDS

2007-2009 Shanghai AVR single chip   programming contest (First prize in all three years)
2009,   2011 "INTEL" Innovation Contest in Shanghai
2009 Future Engineering Contest in Shanghai
2009-2015 Six consecutive years of Shu Ping Scholarship(top 3%)
2012 Fudan University Excellent Freshman Scholarship(top 5%)
2013-2014 Two consecutive years of Fudan university scholarship(top 15%)
2015 Mathematical Contest in Modeling 2015 (Meritorious Winner, top 10%)
2018 GSFEI Travel Funding, University of Washington
2018 Travel Stipend Awards, SMRA
2018 Outstanding Research Award, OCSMRM
2018, 2019, 2020 ISMRM Annual Meeting & Exhibition Stipend
2019, 2020 Magna Cum Laude Merit Award, ISMRM
2019 Winner for the AHA-AWS award at the 2019 American Heart Association Scientific Meeting


ADVISORS

Alvin Chen

Group Leader and Senior Scientist, AI for Healthcare @ Philips
From 2021.10 — Present

Our group at Philips is developing first-in-kind AI-powered imaging and diagnostic technologies to improve healthcare access and emergency response, particularly in point-of-care and resource-limited settings. We focus on applying machine learning techniques and best practices to address challenges in data, annotation, clinical validation, and real-world deployment of these technologies. Specialization: medical image analysis, machine and deep learning, robotics, AI for healthcare, clinical-translational research, academic-industrial partnerships

Email: alvin.chen@philips.com

Experience with me:

             Manager in Philips

Chun Yuan

Professor — Department of Radiology and Bioengineering, University of Washington,
From 2016.9 — 2021.8

Researches on Magnetic Resonance Imaging, Vulnerable Plaque/Vessel Wall Imaging and Analysis, Cardiovascular Disease Analysis and Investigation, Vessel Imaging of Vascular Cognitive Impairment and Dementia, MRI Sequence Development and Image Reconstruction, Vessel Image Processing With AI, Machine Learning, and Deep Learning.

Email: cyuan@uw.edu

Experience with me:

             Course: MRI & ULTRASOUND
             PhD thesis co-chair.

Jenq-Neng Hwang

Professor — Department of Electrical & Computer Engineering University of Washington,
From 2016.9 — 2021.8

Researches on Image/video signal processing, Multimedia network and QoS, Statistical pattern recognition.

Email: hwang@uw.edu

Experience with me:

             PhD thesis co-chair.

Huiliang Shang

Accosiate Professor — Department of Electronic Engineering Fudan University,
From 2013.10 — 2016.6

Researches on image processing, Biomedical topics, Traditional Chinese medicine, Robotic, Computer vision, visible light communication, Circuit theory.

Email: shanghl@fudan.edu.cn

Experience with me:

             Course: Circuit Theory
             Four papers about Biomedical Image processing
             LED position car etc.

Bin Wang

Professor — Department of Electronic Engineering, Fudan University
From 2013.10 — 2014.11

Researches on image processing, signal, pattern recongnition, intelligent information proessing and brain science

Email: wangbin@fudan.edu.cn

Experience with me:

             Course: Probability

             The FDUROP project about image processing

Xiang Li

Professor — Department of Electronic Engineering, Fudan University
From 2014.3 — 2015.3

Researches on complex network and multi-agent systems.

Email: lix@fudan.edu.cn

Experience with me:

             Course: Introduction to Network Science

             Paper about immunization network

Yi Guo

Professor — Department of Electronic Engineering , Fudan University,
From 2015.8 — 2016.6

Researches on Medical image and signal processing, Medical ultrasonics.

Email: guoyi@fudan.edu.cn

Experience with me:

             Course: Medical Imaging
             Capstone research advisor
             One paper on vascualr registration

Yuanyuan Wang

Professor — Department of Electronic Engineering , Fudan University,
From 2015.8 — 2016.6

Researches on Medical image and signal processing, Medical ultrasonics.

Email: yywang@fudan.edu.cn

Experience with me:

             Course: Signal and System
             Final year project research group advisor
             



ACTIVITIES

2012-2013 Volunteer teaching in Shanghai Sunflower Weekend School, English Junior class.
2012-2014 Fudan Student Union, Department of Information. Technical director in 2014.
2014.7 Volunteer teaching in Yunnan. Lecturer in Paper cutting and Handwork (Build a gravity toy car)
2014-2015 Technical group of Shu Ping Scholarship Foundation. Developed online scholarship application system.
2017-2018 Trainee Member, Radiological Society of North America (RSNA)
2015-2021 IEEE student member.
2017-2021 Trainee Member, International Society for Magnetic Resonance in Medicine (ISMRM).
2019-2020 Student member, American Heart Association
2019-2021 Student member, Society for Medical Image Computing and Computer Assisted Intervention.



Comment 评论