Hanxue Gu

hanxue (dot) gu (at) duke (dot) edu

Duke University

Durham, NC, United States, 27703

About me: I am a 5th year Ph.D. student in Electrical and Computer Engineering at Duke University, working at the intersection of AI and Healthcare. I am fortunate to be advised by Prof. Maciej A. Mazurowski under Duke Spark Initiative. My research sits at the intersection of machine learning and healthcare, with a focus on developing and adapting deep learning methods for medical image analysis—from application-oriented tools to foundational advancements.

I am interested in several research topics, including foundation models, self- and semi-supervised learning, image harmonization, and clinically meaningful AI. My work spans segmentation, registration, 2D-to-3D reconstruction, and risk prediction. See my Research Topics Bar for details.

Before starting my Ph.D., I received my Bachelor’s degree from Zhejiang University, where I studied Electrical & Information Engineering and minored in ACEE under the Chu Kochen Honors College. Prior to Duke, I spent a year as a research intern at the Athinoula A. Martinos Center for Biomedical Imaging at Massachusetts General Hospital (MGH), where I worked on optical image reconstruction and deep learning.

See my Google Scholar for a full list of my publications, with a few highlights shown below.

Feel free to connect with me on LinkedIn or contact me through email— I’m always happy to chat! (🤗)

news

Aug 02, 2025	🎉 Thrilled to share that two of our papers on breast MRI have been accepted to the Deep-Breath Workshop at MICCAI 2025! 🧠 The first paper presents the most comprehensive breast MRI segmentation dataset to date, covering a wide range of anatomical structures. We hope this effort can lay the foundation for future research in breast composition analysis, risk stratification, and AI-driven screening tools. 🌀 The second paper focuses on generalizable breast MRI registration, addressing the challenge of aligning highly deformable anatomy across patients and imaging sessions. We demonstrate the power of foundation model features and training-free registration strategies to enable robust alignment without manual annotation. We believe these works together push toward more generalizable, scalable, and interpretable tools for breast MRI analysis. 🚀 Stay tuned — datasets, code, and more resources will be released soon. Big thanks to the Deep-Breath organizers and all our co-authors and collaborators!
Jun 29, 2025	🩺 Excited to share that our paper “Breast density in MRI: an AI-based quantification and relationship to assessment in mammography” has been accepted for publication in npj Breast Cancer! In this study, we developed an AI-driven pipeline to quantify breast density directly from MRI scans and explored how these MRI-derived metrics relate to traditional density assessments from mammography. Our work provides new insights into leveraging MRI for more comprehensive breast density evaluation. Huge thanks to all co-authors and collaborators — looking forward to seeing this work published soon!
Jun 29, 2025	🎉 Excited to share that the paper I contributed as co-author “SAMora: Enhancing SAM through Hierarchical Self-Supervised Pre-Training for Medical Images” has been accepted to ICCV 2025! In this work, we propose SAMora, a novel framework that boosts the performance of the Segment Anything Model (SAM) on medical images through a hierarchical self-supervised pre-training strategy. Our approach brings significant improvements to medical segmentation tasks where labeled data is scarce. Thank you to all co-authors and collaborators for making this possible！
Jun 25, 2025	🧠 Excited to share our latest work “MRI-CORE: A Foundation Model for Magnetic Resonance Imaging”, now available on arXiv! This project introduces a general-purpose MRI foundation model capable of handling diverse MRI sequences and anatomical regions, laying the groundwork for improved downstream tasks across medical imaging applications. 📄 Read the paper on arXiv Thank you to all collaborators for making this happen!
Jun 12, 2025	🎉 Our paper “Deep Learning Automates Cobb Angle Measurement Compared with Multi-Expert Observers” has been accepted to the British Journal of Radiology: Artificial Intelligence! This work presents a clinically interpretable AI model for scoliosis assessment, validated in a multi-reader study against experienced clinicians. Big thanks to our collaborators!

selected publications

How to build the best medical image segmentation algorithm using foundation models: a comprehensive empirical study with Segment Anything Model

Hanxue Gu, Haoyu Dong, Jichen Yang, and 1 more author

Machine Learning for Biomedical Imaging, May 2025

Abs DOI Bib Code

@article{Gu_2025,
  title = {How to build the best medical image segmentation algorithm using foundation models: a comprehensive empirical study with Segment Anything Model},
  volume = {3},
  issn = {2766-905X},
  url = {http://dx.doi.org/10.59275/j.melba.2025-86a6},
  doi = {10.59275/j.melba.2025-86a6},
  number = {May 2025},
  journal = {Machine Learning for Biomedical Imaging},
  publisher = {Machine Learning for Biomedical Imaging},
  author = {Gu, Hanxue and Dong, Haoyu and Yang, Jichen and Mazurowski, Maciej A.},
  year = {2025},
  month = may,
  pages = {88–120},
  dimensions = {true},
  bibtex_show = true,
}

Segmentanybone: A universal model that segments any bone at any location on mri

Hanxue Gu, Roy Colglazier, Haoyu Dong, and 8 more authors

Medical Image Analysis, May 2025

Abs Bib Code

@article{gu2025segmentanybone,
  title = {Segmentanybone: A universal model that segments any bone at any location on mri},
  author = {Gu, Hanxue and Colglazier, Roy and Dong, Haoyu and Zhang, Jikai and Chen, Yaqian and Yildiz, Zafer and Chen, Yuwen and Li, Lin and Yang, Jichen and Willhite, Jay and others},
  journal = {Medical Image Analysis},
  pages = {103469},
  year = {2025},
  publisher = {Elsevier},
  dimensions = {true},
  bibtex_show = true,
}

SuperMask: Generating High-resolution object masks from multi-view, unaligned low-resolution MRIs

Hanxue Gu, Hongyu He, Roy Colglazier, and 3 more authors

In Medical Imaging with Deep Learning, May 2024

Abs Bib

@inproceedings{gu2024supermask,
  title = {SuperMask: Generating High-resolution object masks from multi-view, unaligned low-resolution MRIs},
  author = {Gu, Hanxue and He, Hongyu and Colglazier, Roy and Axelrod, Jordan and French, Robert and Mazurowski, Maciej A},
  booktitle = {Medical Imaging with Deep Learning},
  pages = {119--133},
  year = {2024},
  organization = {PMLR},
  dimensions = {true},
  bibtex_show = true
}

Improving Surgical Risk Prediction Through Integrating Automated Body Composition Analysis: a Retrospective Trial on Colectomy Surgery

Hanxue Gu, Yaqian Chen, Diego Schaps, and 5 more authors

arXiv preprint arXiv:2506.11996, May 2025

Abs Bib

@article{gu2025improving,
  title = {Improving Surgical Risk Prediction Through Integrating Automated Body Composition Analysis: a Retrospective Trial on Colectomy Surgery},
  author = {Gu, Hanxue and Chen, Yaqian and Schaps, Diego and Woody, Regina and Colglazier, Roy and Mazurowski, Maciej A and Mantyh, Christopher and others},
  journal = {arXiv preprint arXiv:2506.11996},
  year = {2025},
  dimensions = {true},
  bibtex_show = true
}

Segment Anything Model for Medical Image Analysis: An Experimental Study

Maciej A Mazurowski, Haoyu Dong, Hanxue Gu, and 3 more authors

Medical Image Analysis, May 2023

Abs Bib

@article{mazurowski2023segment,
  title = {Segment Anything Model for Medical Image Analysis: An Experimental Study},
  author = {Mazurowski, Maciej A and Dong, Haoyu and Gu, Hanxue and Yang, Jichen and Konz, Nicholas and Zhang, Yixin},
  journal = {Medical Image Analysis},
  volume = {89},
  pages = {102918},
  year = {2023},
  dimensions = {true},
  bibtex_show = true
}

Segment Anything Model 2: An Application to 2D and 3D Medical Images

Haoyu Dong^*, Hanxue Gu^*, Yaqian Chen, and 3 more authors

arXiv preprint arXiv:2408.00756, May 2024

Abs Bib

We extend the Segment Anything Model (SAM2) to handle both 2D and 3D medical imaging tasks, demonstrating significant improvements in segmentation flexibility and performance across diverse datasets.
@article{dong2024segment, title = {Segment Anything Model 2: An Application to 2D and 3D Medical Images}, author = {Dong, Haoyu and Gu, Hanxue and Chen, Yaqian and Yang, Jichen and Chen, Yuwen and Mazurowski, Maciej A}, journal = {arXiv preprint arXiv:2408.00756}, year = {2024}, dimensions = {true}, bibtex_show = true }

A publicly available deep learning model and dataset for segmentation of breast, fibroglandular tissue, and vessels in breast MRI

Christopher O Lew, Majid Harouni, Ella R Kirksey, and 7 more authors

Scientific Reports, May 2024

Abs Bib Code

@article{lew2024publicly,
  title = {A publicly available deep learning model and dataset for segmentation of breast, fibroglandular tissue, and vessels in breast MRI},
  author = {Lew, Christopher O and Harouni, Majid and Kirksey, Ella R and Kang, Elianne J and Dong, Haoyu and Gu, Hanxue and Grimm, Lars J and Walsh, Ruth and Lowell, Dorothy A and Mazurowski, Maciej A},
  journal = {Scientific Reports},
  volume = {14},
  number = {1},
  pages = {5383},
  year = {2024},
  publisher = {Nature Publishing Group UK London},
  dimensions = {true},
  bibtex_show = true,
}

FDU-net: deep learning-based three-dimensional diffuse optical image reconstruction

Bin Deng^*, Hanxue Gu^*, Hongmin Zhu, and 5 more authors

IEEE transactions on medical imaging, May 2023

Abs Bib

@article{deng2023fdu,
  title = {FDU-net: deep learning-based three-dimensional diffuse optical image reconstruction},
  author = {Deng, Bin and Gu, Hanxue and Zhu, Hongmin and Chang, Ken and Hoebel, Katharina V and Patel, Jay B and Kalpathy-Cramer, Jayashree and Carp, Stefan A},
  journal = {IEEE transactions on medical imaging},
  volume = {42},
  number = {8},
  pages = {2439--2450},
  year = {2023},
  publisher = {IEEE},
  dimensions = {true},
  bibtex_show = true
}

The intrinsic manifolds of radiological images and their role in deep learning

Nicholas Konz, Hanxue Gu, Haoyu Dong, and 1 more author

In International Conference on Medical Image Computing and Computer-Assisted Intervention, May 2022

Abs Bib

We compare the intrinsic manifold structure of radiological and natural images, finding that although radiological images have lower intrinsic dimensionality, their features are harder to learn, which helps explain why applying deep learning to medical images remains more challenging than to natural image datasets.
@inproceedings{konz2022intrinsic, title = {The intrinsic manifolds of radiological images and their role in deep learning}, author = {Konz, Nicholas and Gu, Hanxue and Dong, Haoyu and Mazurowski, Maciej A}, booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention}, pages = {684--694}, year = {2022}, bibtex_show = true, dimensions = {true}, organization = {Springer Nature Switzerland Cham} }

Medical image segmentation with intent: Integrated entropy weighting for single image test-time adaptation

Haoyu Dong, Nicholas Konz, Hanxue Gu, and 1 more author

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, May 2024

Bib

@inproceedings{dong2024medical,
  title = {Medical image segmentation with intent: Integrated entropy weighting for single image test-time adaptation},
  author = {Dong, Haoyu and Konz, Nicholas and Gu, Hanxue and Mazurowski, Maciej A},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages = {5046--5055},
  year = {2024},
  dimensions = {true},
  bibtex_show = true,
}