Research
Our work spans five connected fronts, each aimed at moving AI from the lab bench to the bedside.
- Computer Vision We develop novel vision architectures — including Swin-Transformer-based models for image super-resolution and segmentation — for high-stakes visual recognition tasks where performance and reliability are non-negotiable, from reconstructing degraded medical scans to detecting sub-millimeter lesions.
- Clinical AI & Medical Imaging Our imaging research spans OCT, fundus, mammography, MRI, PET, and CT. Past work includes NASA-funded super-resolution imaging to detect Spaceflight-Associated Neuro-Ocular Syndrome (SANS) in astronauts, and NSF-funded segmentation of micro-scale breast masses that conventional methods miss.
- Deep Learning & Machine Learning We investigate knowledge distillation and multimodal learning — including Teach-Former, a multi-teacher framework that lets lightweight student models learn jointly from CT, PET, and MRI — to build models accurate enough for the clinic and light enough for resource-constrained settings.
- Patient Outcomes & Clinical Research We partner with clinicians to measure how AI tools affect real patient outcomes, workflow efficiency, and diagnostic accuracy, ensuring our research translates into meaningful clinical benefit.
- Trustworthy AI in Healthcare We study explainability, uncertainty quantification, bias detection, and fairness in clinical AI systems, including adversarial robustness — our GAN-based defenses and game-theoretic attack models protect diagnostic systems from adversarial threats.
Looking ahead, we’re extending this work toward foundation models and large language models for multimodal diagnostic reporting, bridging AI output and clinical insight.