Research
We build AI methods that work with the messy, multimodal data of real clinical practice — signals, images, video and text — in close partnership with physicians at UC Davis Health.
- Multimodal AI
- Generative AI & LLMs
- Synthetic data
- Retrieval-augmented generation
- Medical imaging
Research areas
Cardiac arrhythmia and atrial fibrillation
We model the recurrence of irregular heart rhythms following treatment, combining clinical, imaging and signal data into multimodal predictive models. Recent work addresses bias and missing modalities in medical multimodal representation learning.
Medical imaging and interventional radiology
We develop deep learning models that identify abnormalities and anatomy in radiological images and video, including automatic classification and hierarchical localization of vascular structures in digital subtraction angiography, and report generation for mammography.
Large language models for clinical documents
We study how LLMs can reliably annotate, summarize and generate clinical text while keeping data private. Projects include LLM-based radiology report annotation (RadAnnotate), local multi-model RAG pipelines for mammography reports (MammoWise), automating procedure case logging in medical education, and mining biomedical literature for protein–disease associations.
Precision and population health
We apply data science and NLP to understand broader determinants of health — from the social exposome in cardiovascular disease, to the mental health needs of Alzheimer's caregivers expressed in online forums, to forecasting COVID-19 hospital admissions with UC Davis Medical Center.
Funding and partners
Our work in AI for Health has been supported by the National Institutes of Health (NIH) and the UC Davis Clinical and Translational Science Center (CTSA). We are part of the CS AI for Health Initiative and work with UC Davis DataLab, the UC Davis Health Center for Precision Medicine, and UC Davis Grand Challenges.