Ben Rush
Imaging informatics
Research at the intersection of routine imaging, automated biomarker extraction, survival prediction, and safer radiology AI workflows.
The throughline is rigorous translation: harmonized cohorts, site-level quality control, cautious prediction language, and models that are useful because they are understandable.
Benjamin Rush, PhD, MPH is an informatics data scientist in the Department of Radiology at the University of Wisconsin–Madison whose research integrates CT-based biomarkers, large-scale imaging cohorts, and artificial intelligence to improve disease risk stratification and population health insight.
His work spans automated body composition analysis from abdominal CT, multi-site harmonization of imaging data, survival modeling, and the implementation of machine learning and large language models within radiology report workflows.
With training in epidemiology and nutritional sciences, he brings a translational lens to imaging informatics—connecting physiology, muscle and adipose tissue biology, and socioeconomic context to clinically actionable imaging biomarkers.
Across projects involving datasets bordering one million scans, he emphasizes reproducible pipelines, center-scale collaboration, and clear, empathetic science communication, operating under the philosophy that “with function comes beauty” in both computational systems and scientific discovery.
Full publication record on ORCID.
For scientific collaboration: rush4@wisc.edu
Radiology AI
I study how large language models can help identify incidental breast findings in radiology reports, with a focus on prompt design, model agreement, sensitivity, specificity, and consensus methods that could support safer follow-up workflows.
OSCAR
I work on harmonizing opportunistic abdominal CT biomarkers across participating sites to evaluate generalizability, site-level robustness, and population-level prediction of survival and other outcomes. Biomarkers include L3 muscle area and density, SAT/VAT area, VAT density, visceral-to-subcutaneous fat ratio, L1 bone attenuation, and aortic calcium.
Public health
I link CT-derived biomarkers with neighborhood-level socioeconomic indicators and healthcare access measures to study disparities and community-level patterns in imaging-derived health information.
Body composition
I study muscle quantity, muscle density, adipose tissue distribution, bone attenuation, and longitudinal body composition change across clinical and population-level cohorts.
I care about the parts of clinical data science that make models trustworthy: clear definitions, auditable data cleaning, appropriate statistics, and interpretation that matches the evidence.
Cohorts
Define the index scan, deduplicate patients when needed, harmonize dates, align clinical variables, and preserve row-count audit trails through every major filtering step.
QC
Inspect medians, IQRs, ranges, missingness, implausible values, extraction failures, scanner/protocol variation, and site-level differences before modeling.
Modeling
Use survival and fixed-horizon prediction approaches when appropriate, reporting discrimination, calibration, uncertainty, and clinically relevant operating characteristics.
Interpretation
Frame findings as prediction, association, risk stratification, transportability, or external validation unless a diagnostic or causal claim has been specifically validated.
Journal of Clinical Densitometry · 2026
American Journal of Roentgenology · 2025
JCSM Communications · 2024
Journal of Cachexia, Sarcopenia and Muscle · 2022
Dissertation · 2022
Frontiers in Rehabilitation Sciences · 2022
JBMR Plus · 2021
The FASEB Journal · 2020
Journal of Immigrant and Minority Health · 2016