Peng Liu
Iowa State University, IA·Mathematics & Statistics
Large biomedical imaging databases make it possible to ask whether a treatment or exposure changes what shows up in an image, such as a brain scan. Answering such cause-and-effect questions from observational data is hard.
- Fields
Mathematics & Statistics
Computer Science
Public Health & Epidemiology
- Topics
functional data analysis
causal inference
medical imaging
- Grants
- 1 known grant, through Aug 2028
What they work on
- Develop functional regression models for imaging outcomes built on a distributed learning framework, so that estimation and inference stay accurate yet scalable for large imaging datasets.
- Introduce an image-on-scalar instrumental variable regression method to reduce bias from unmeasured confounding when estimating treatment effects on images in observational studies.
- Propose an image-on-scalar doubly robust regression method using functional pseudo-outcomes to handle population heterogeneity, then evaluate all methods on existing imaging studies and release open-source software.
More about the research
The images are huge. Hidden factors can distort results, a problem called unmeasured confounding. And different people may respond differently, which is called population heterogeneity. Existing statistical methods do not handle these problems well at large scale.
Why it matters
Better tools for drawing causal conclusions from large observational imaging studies could help public health researchers understand how treatments affect the body without running a new trial. The free software is meant for researchers using public imaging repositories. It could also apply to aging research, digital health, and plant science.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Yang Ni
University of Texas at Austin, TX·Mathematics & Statistics
Develop a Bayesian joint model of treatment, outcome, covariates, and hidden confounders that identifies causal effects from observational data without instrumental or proxy variables.
Medellena Glymour
Boston University, MA·Public Health & Epidemiology
Estimate how low-risk drinking, unhealthy drinking and alcohol use disorder each affect dementia risk, triangulating across ten cohorts, including Kaiser Permanente records and the UK Biobank, using doubly robust and instrumental variable methods.
Matias Cattaneo
Princeton University, NJ·Mathematics & Statistics
Demonstrate the empirical and theoretical limitations of binned regression models applied to mixed-frequency panel data in climate economics, showing when they can lead to incorrect conclusions.
The school
Iowa State University: 23 labs on PI Pages, $462M a year on research, 327 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.