Human-Centered & Translational AI for Medicine
I'm a
Quaecumque sunt vera — Whatsoever things are true
I
How can medical AI learn from the clinicians it works beside, and grow into a better partner with every case?
I am a PhD student at Northwestern University, working across the McCormick School of Engineering and the Feinberg School of Medicine, advised by Prof. Ulas Bagci in the Machine & Hybrid Intelligence Lab. Before Northwestern, I was a Research Associate at Carnegie Mellon University, and I earned my MSc in AI in Health from Université Bourgogne Europe, graduating as Salutatorian.
My research follows two connected themes:
Dynamic, self-improving (recursive) AI that adapts to human behavior across modalities: what clinicians see, say, look at and think, from vision and language to eye gaze and brain signals.
Bringing those systems to clinicians and radiologists, to work side by side on diagnosis, prognosis and treatment across MRI, CT, PET and other modalities, and solving the research problems in the way: 3D segmentation, label scarcity, domain generalization.
I also founded AIMA Research, which has brought 100+ Vietnamese students into Health AI, and I have taught with AI VIETNAM since 2022. I'm always happy to talk about ideas and collaborations.
II
Human-centered AI that works side by side with clinicians. I design dynamic, self-improving AI that learns from how people see, speak, look and think, and I carry it into the clinic for diagnosis, prognosis and treatment across MRI, CT, PET and beyond. Select any label in the figure.
Dynamic AI does not freeze after training. It keeps learning from the people who use it: refining its own outputs, updating as experts correct them, and adapting to how each clinician works. My teacher–student and self-correcting frameworks are early forms of this loop.
Representative: Switching Dual-Student framework (Pattern Recognition 2026) · Scribble-supervised segmentation with dynamic teacher switching (ISBI 2026 Oral) · ESC: Emotional Self-Correction for VLMs (ECCV 2026)
What the clinician sees is the first input. I work across MRI, CT, ultrasound, chest X-ray, mammography, digital pathology and cryo-electron tomography, so that models learn from the same images a radiologist reads.
Representative: DuetMatch: brain MRI segmentation (CMIG 2025) · Unsupervised cryo-ET segmentation (CVPR 2026 Highlight)
What clinicians say and write: reports, prompts and questions. Vision-language models that describe a region, answer a question about an image, and correct their own mistakes.
Representative: Describe Anything in Medical Images (ICML 2025 Workshop) · ESC: Emotional Self-Correction (ECCV 2026)
Where the radiologist looks while reading a scan is a natural, effortless signal of attention. Gaze can tell a model what matters in an image and let it adapt to a reader in real time, without any extra annotation.
How the clinician thinks. Brain signals recorded while reading images are a feedback signal for AI that adapts to human cognition: when a reader is uncertain, surprised, or confident.
Detection, classification and screening side by side with the reader: long-tailed chest X-ray findings, breast lesions in mammography and ultrasound, and the question of which patient subgroups a model misses.
Representative: Long-tailed & zero-shot chest X-ray classification (ISBI 2026 Oral) · Who Gets Missed in the Tail? (Medical Image Analysis)
Predicting risk and outcomes from imaging: radiomics signatures for survival, and imaging anchored to molecular data.
Representative: CT radiomics for risk stratification in colorectal liver metastases (Physics in Medicine & Biology) · Imaging-anchored multiomics in cardiovascular disease (Briefings in Bioinformatics)
Segmentation and staging that feed treatment planning: liver in multi-phase MRI, brain tumors, fetal ultrasound, and laryngeal cancer staging.
Representative: Cross-modality liver segmentation in multi-phase MRI (MICCAI 2025 Workshop) · Semi-supervised multi-modal brain tumor segmentation (AAAI 2026 Bridge)
Volumetric segmentation of organs, tumors and cells in 3D, from multi-phase liver MRI to cryo-electron tomograms and 3D spheroid cultures.
Representative: Blurry-Consistency 3D spheroid segmentation (CVPR 2024 Workshop) · Multi-modal brain tumor segmentation (AAAI 2026 Bridge)
Expert annotation is the bottleneck. Semi-, weakly- and scribble-supervised learning, mixture-of-experts and pseudo-label denoising approach fully supervised accuracy from a fraction of the labels.
Representative: Semi-MoE (BMVC 2025) · Switching Dual-Student framework (Pattern Recognition 2026)
Models trained at one hospital often fail at the next. Source-free and mixed-domain adaptation keep models working on new scanners, protocols and modalities without the original training data.
Representative: UP2D (Neurocomputing 2026) · Denoised Patch Mixing (ICASSP 2026)
Adapting the Segment Anything Model to unlabeled and source-free medical images, bringing region-level description to medical imaging, and making vision-language models reliable.
Representative: From Specialist to Generalist: SAM on unlabeled images (ISBI 2026 Oral) · Adapting SAM Without Labels (ICIP 2026)
Dynamic, self-improving (recursive) AI that adapts to human behavior across modalities.
Bringing those designs to clinicians and radiologists, to work side by side on every task.
The open research problems that make translation possible.
Rotate, slice and explore real MRI and CT volumes. Loads about 3.7 MB on demand.
The kind of data my models work with every day: brain MRI, contrast-enhanced MRI and head CT, in full 3D. Rotate the rendering, or switch to slices and read them the way a radiologist does. A segmentation model has to hold up across all of them, whatever the scanner, contrast or anatomy.
Tap two nuclei to teach the model
Label two cell nuclei and the model segments the rest. Tap any mistake to correct it: each correction changes what the model looks for, and it re-segments on the spot. A small version of the human-in-the-loop, self-improving AI I build for clinicians.
Simulated fluorescence microscopy, generated in your browser, so the true nuclei are known and every correction can be scored.
III
Northwestern University — McCormick School of Engineering & Feinberg School of Medicine
Université Bourgogne Europe, France
Machine & Hybrid Intelligence Lab · Prof. Ulas Bagci
Computational Biology Department · Prof. Min Xu
CT Liver Segmentation & Registration team, Vietnam
FPT Camera — Face Recognition, Vietnam
Université de Bourgogne Europe, France
Région Bourgogne-Franche-Comté, France
UBFC, France
UBFC, France
National Science and Technology Council, Taiwan
Ministry of Education, Taiwan
Vietnam
A mentorship community I founded to bring Vietnamese undergraduate and master’s students into Health AI research, alongside faculty and mentors across the US.
IV
50+ papers at venues including CVPR, ECCV, ICCV, WACV, BMVC, MICCAI, MIDL, ISBI, ICASSP and AAAI, and in journals such as Medical Image Analysis, npj Digital Medicine, Pattern Recognition and IEEE JBHI.
A switching dual-student framework that adaptively transfers knowledge between students.
Read paperAdapts a trained segmenter to a new domain without source data by progressively cleaning its own pseudo-labels.
Read paperSegmentation from sparse scribble annotations with dynamic teacher switching and hierarchical consistency.
Read paperSwipe for more
* Equal contribution
ICIP 2026 IEEE International Conference on Image Processing Paper
ECCV 2026 European Conference on Computer Vision Paper
ICASSP 2026 International Conference on Acoustics, Speech, and Signal Processing Co-first author Paper
ICASSP 2026 International Conference on Acoustics, Speech, and Signal Processing Co-first author Paper
CVPR 2026 Conference on Computer Vision and Pattern Recognition Findings Paper
CVPR 2026 Conference on Computer Vision and Pattern Recognition Highlight Paper
AAAI 2026 AIMedHealth Bridge Program Workshop Paper
ACDSA 2026 International Conference on Artificial Intelligence, Computer, Data Sciences, and Applications Paper
WACV 2026 Winter Conference on Applications of Computer Vision Paper
MIDL 2026 Medical Imaging with Deep Learning Paper
MIDL 2026 Medical Imaging with Deep Learning Paper
IEEE ISBI 2026 International Symposium on Biomedical Imaging First author Oral Paper
IEEE ISBI 2026 International Symposium on Biomedical Imaging Oral Paper
IEEE ISBI 2026 International Symposium on Biomedical Imaging Oral Paper
IEEE ISBI 2026 International Symposium on Biomedical Imaging Oral Paper
IEEE ISBI 2026 International Symposium on Biomedical Imaging Oral Paper
Medical Image Analysis 2026 IF 11.8 Paper
The of Thoracic and Cardiovascular Surgery 2026 IF 4.7
Pattern Recognition 2026 Co-first author IF 7.6 Paper
npj Digital Medicine 2026 IF 12.4 Paper
Neurocomputing 2026 Co-first author IF 6.5 Paper
Briefings in Bioinformatics 2026 IF 7.3 Paper
Physics in Medicine & Biology 2026 IF 3.4 Paper
Tropical Medicine and Infectious Disease 2026 First author IF 3.1 Paper
Biomedical Signal Processing and Control 2026 First author IF 4.9 Paper
MICCAI 2025 Emerging LLM/LMM Applications in Medical Imaging Workshop Workshop Paper
MICCAI 2025 Emerging LLM/LMM Applications in Medical Imaging Workshop Workshop Paper
BMVC 2025 British Machine Vision Conference Co-first author Paper
AAAI 2025 AIMedHealth Bridge Program Co-first author Workshop Paper
MIDL 2025 Medical Imaging with Deep Learning Co-first author Paper
MEDINFO 2025 Studies in Health Technology and Informatics Paper
IEEE ICIP 2025 International Conference on Image Processing Paper
CVPR 2025 Precognition Workshop Workshop Paper
ICML 2025 Multi-modal Foundation Models and Large Language Models for Life Sciences Workshop Co-first author Workshop Paper
IEEE ISBI 2025 International Symposium on Biomedical Imaging Paper
ICCV 2025 Computer Vision Systems for Document Analysis and Recognition Workshop Workshop Paper
IEEE IPAS 2025 Image Processing, Applications and Systems Paper
MICCAI 2025 Comprehensive Analysis & Computing of Real-world Medical Images Workshop Workshop Paper
MICCAI 2025 Computational Pathology and Multimodal Data Workshop Workshop Paper
MICCAI 2025 Fast, Low-resource, Accurate, Robust, and Effectual Medical Image Analysis Workshop Workshop Paper
IEEE of Biomedical and Health Informatics 2025 IF 6.7 Paper
Computerized Medical Imaging and Graphics 2025 Co-first author IF 4.9 Paper
Journal of Medical System 2025 IF 5.7 Paper
CVPR 2024 Domain adaptation, Explainability, Fairness in AI for Medical Image Analysis Workshop First author Workshop Paper
IEEE ISBI 2024 International Symposium on Biomedical Imaging Paper
CVPR 2024 Affective and Behavior Analysis in-the-wild Workshop Workshop Paper
IEEE Access 2024 IF 3.4 Paper
ICCV 2023 Computer Vision for Automated Medical Diagnosis Workshop First author Workshop Paper
IEEE SSP 2023 Statistical Signal Processing First author Paper
MICCAI 2023 Machine Learning in Medical Imaging Workshop Co-first author Workshop Paper
No publications match these filters.
V
VI
Institutions across the United States I work with, through joint papers, mentoring and shared projects.
VII
Interested in collaborating, or a student who would like to work on these topics? Get in touch.