Geoffrey Tison
University of California-San Francisco, CA·Computer Science·Early-career
Doctors rarely decide based on a single test. They combine images, recordings and other data to reach a diagnosis.
- Fields
Computer Science
Medicine & Clinical Research
Biomedical Engineering
- Topics
artificial intelligence
deep learning
deep neural networks
- Grants
- 1 known grant, last ended Aug 2026
What they work on
- Develop a deep neural network architecture that accepts multiple raw medical data modalities and incorporates prior physiological knowledge, so it can combine information the way physicians do.
- Build an automated, modality-agnostic pipeline to pre-process and harmonize real-world medical data collected on varied hardware across healthcare institutions.
- Apply and prospectively validate the pre-processing pipeline and multi-modal neural network to identify heart failure-related phenotypes in a large multi-modality cohort of heart failure patients.
More about the research
Most medical artificial intelligence (AI) systems, built on deep neural networks (DNNs), analyze only one type of test at a time. Disease, however, often spans several organ systems. This limits how complex a medical problem AI can address. Two practical barriers stand in the way: designing networks that combine several data types, and cleaning and harmonizing real-world medical data from many device makers and hospitals.
Why it matters
A platform that lets AI combine several kinds of medical tests could improve performance on existing clinical tasks and widen the range of problems AI can help with across medical specialties. In heart failure, it could help identify clinically relevant patient patterns, and it may support data-driven discovery of disease physiology.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Michael Sjoding
University of Michigan, MI·Medicine & Clinical Research
Evaluate computational strategies that make an AI tool for diagnosing acute dyspnea (shortness of breath) more robust and less biased when analyzing clinical and imaging data from different patients.
Qiuying Sha
Michigan Technological University, MI·Computer Science
Identify subgroups of heart failure patients more likely to respond to cardiac resynchronization therapy by applying unsupervised machine learning to combined ECG electrical timing and gated SPECT mechanical timing measures.
Oguz Akbilgic
Wake Forest University, NC·Medicine & Clinical Research
Develop an AI model that predicts future heart failure with preserved ejection fraction (HFpEF) from ECG data, using carefully verified data from NIH-funded studies.
The school
University of California-San Francisco: 89 labs on PI Pages, $2.1B a year on research, 164 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.