Mathias Unberath
Johns Hopkins University, MD·Biomedical Engineering·Early-career
Computers increasingly help guide medical care, from AI that reads X-rays to robots that assist surgeons. These tools often perform worse in real hospitals than in controlled tests, and building them is slow and costly.
- Fields
Biomedical Engineering
Computer Science
Engineering
- Topics
medical AI
machine learning
dataset bias
- Grants
- 3 known grants, through Apr 2031
What they work on
- Build a framework that deliberately adds bias, shortcuts and imaging-condition changes to curated chest X-ray, CT and dermoscopy datasets, to validate methods for auditing medical imaging datasets.
- Develop causality-driven auditing methods that find image signatures linked to factors spuriously tied to outcomes (shortcuts) that would make medical AI models perform poorly.
- Assess how image quality across a dataset affects AI robustness, using causal models of image formation to identify the imaging factors that limit performance.
- Develop and test a low-profile, cannula-mounted piezo robot on a passive positioning arm that is accurate and strong enough for vertebral augmentation without blocking X-ray guidance.
- Create algorithms for a camera-augmented C-arm X-ray system that automate planning and guidance, deliver them through mixed reality, and validate the full system in cadaver studies.
- Develop computer vision and machine learning methods to build digital twins of surgical environments, and use them to speed surgical robot design and improve situational awareness.
More about the research
The lab's projects ask how to make imaging-driven medical technology more trustworthy and practical. One project audits the datasets that medical AI learns from. The second develops small, affordable robots guided by X-ray images for spine procedures. The third builds "digital twins," detailed virtual copies of surgical environments.
Why it matters
Dataset audits could help developers and regulators of AI medical software catch biases before deployment, reducing unfair or unreliable results for patients. Low-cost, tool-mounted robots could bring safer, lower-radiation vertebral augmentation to people with osteoporosis- or cancer-related spinal fractures. Digital twins could make designing surgical robots faster and cheaper, and make surgery safer.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Quantitative dataset audit methods to identify bias, shortcut, and robustness risks for data-driven healthcare AI
NIH · Jul 2026 – Apr 2031
CAREER: Digital Twins of Surgical Environments for Situational Awareness and Immersive Simulation
NSF · Jul 2023 – Jun 2028
A Multi-robot System for Semi-automated Image-guided Vertebral Augmentation
NIH · May 2024 – Feb 2028
Closest labs
Ronald Stanley
Missouri University of Science and Technology, MO·Engineering
Build an online platform with the International Skin Imaging Collaborative (ISIC) where experts worldwide annotate, segment and validate critical structures in dermoscopic skin lesion images.
Mitsunori Ogihara
University of Miami, FL·Computer Science
Develop accelerated methods for analyzing medical images, using AI/ML together with coupled software and programmable hardware so results arrive fast enough for real-time use.
William A Grissom
Case Western Reserve University, OH·Biomedical Engineering
Develop fast intraoperative imaging on a low-field MRI scanner so the surgeon's image-guidance display updates in near real time as prostate tissue is resected.
The school
Johns Hopkins University: 100 labs on PI Pages, $4.1B a year on research, 622 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.