About Me
Hello! I am Haodong Li, a dedicated researcher and engineer with a strong background in Electrical and Computer Engineering. My passion lies in advancing artificial intelligence and deep generative models to solve challenging inverse problems in medical imaging and precision healthcare.
I strive to merge cutting-edge research with practical applications, making significant contributions in fields such as CT image reconstruction, segmentation of medical images, and the integration of large language models with advanced vision tasks.
Research & Projects
Research Interests
I focus on probabilistic methods in machine learning and the development of deep generative models to address inverse problems. My research highlights include:
- Designing generative models that capture uncertainty and structure in real-world data.
- Developing efficient diffusion transformers for clinical few-view CT reconstruction.
- Leveraging large language models for innovative medical AI image applications.
CDPIR: Cross-Distribution Diffusion Priors-Driven Iterative Reconstruction for Sparse-View CT
First‑author journal paper — under review at IEEE Transactions on Medical Imaging (TMI)
What we propose. A domain‑adaptive reconstruction framework that marries diffusion‑transformer priors with physics‑based iterative updates. The model disentangles domain‑invariant priors for anatomy and domain‑specific priors for texture/edges, then alternates denoising with data‑consistency to suppress streak artifacts and preserve details under distribution shifts.
- Method: SiT‑based velocity/score sampling under a unified stochastic interpolant; classifier‑free guidance; residual‑guided alternating solver with improved ASD‑POCS.
- Datasets: AAPM Low‑Dose CT, Stanford COCA, XCAT phantom, GE clinical cardiac, and MARS PCCT extremity.
- Results (highlights): State‑of‑the‑art OOD performance; +2–3 dB PSNR and up to +0.10 SSIM vs. baselines across OOD tests; +0.07 SSIM on GE clinical cardiac (zero‑shot); about +3.5 dB PSNR and +0.05 SSIM on MARS PCCT; competitive quality with as few as 200 sampling steps.
Research Experience
-
Imaging and Informatics Lab, University of Massachusetts Lowell (09/2023 – Present)
- Developing state-of-the-art diffusion models to reconstruct high-quality CT images from limited-angle scans.
- Integrating large language models with diffusion models to guide vision tasks using text prompts.
-
Shao’s Lab, University of Florida (05/2023 – 08/2023)
- Developed an AI model for automatic segmentation of the cardiac chambers on CMR images.
- Built a predictive AI model to identify radiomic signatures of cardiac sarcoidosis.
-
Computer Vision and Sensing Systems (COVISS) Lab, University of Florida (08/2022 – 05/2023)
- Employed unsupervised learning (SimCLR) to enhance transformer-based connectivity across different camera views.
- Developed a transformer model for tracking individuals to improve airport safety.
-
Xidian University (09/2020 – 06/2021)
- Proposed an unpaired training method using coupled-GANs for low-dose CT denoising.
Professional Experience
- UF ABE Robotics Team (06/2022 – 08/2022)
• Led the computer vision segment of the robotics project, extracting 3D point clouds and implementing YOLO for precise detection, culminating in a third-place finish in the ASABE 2022 competition.
Technical Skills
Programming: Python, C, C++, Matlab, VHDL
Frameworks: TensorFlow, PyTorch
Software: Multisim, Quartus, Qt, Adobe Photoshop, 3D Slicer
Other: Linux, Cloud Computing
Education
Doctor of Engineering in Electrical Engineering
University of Massachusetts Lowell (09/2023 – Present)
Advisor: Hengyong Yu, FIEEE
Master of Science in Electrical and Computer Engineering
University of Florida (08/2021 – 08/2023)
GPA: 3.76/4.0
Bachelor of Engineering in Electronic and Information Engineering
Xidian University (09/2017 – 07/2021)
GPA: 3.5/4.0
Award: Outstanding Learner Scholarship (Top 5%)