My research interests are optimal transport learning for biomedical imaging analysis and medical computer vision. In addition, I am very interested in smart and connected health. I got my PhD from Imaging and Data Science Lab under Prof. Gustavo Rohde guidance, University of Virginia. Previously, I was a research associate @ ESC Lab under Prof. Wenyao Xu guidance from 2014-2015, University at Buffalo. I got my M.S. degree from University at Buffalo.
Left panel: few-shot illumination-invariant face recognition using sliced-Wasserstein metric; middle panel: image classification (top) and time-series classification (bottom) with limited training data using optimal transport; right panel: brain hemorrhage segmentation and image-based single-cell platelet aggregates profiling.
We develop several subspace learning models in sliced-Wasserstein space by leveraging certain mathematical properties of the Radon Cumulative Distribution Transform (R-CDT) to perform data-efficient learning, such as illumination-invariant face recognition and image classification with limited training data.
In addition, we collaborate with scholars from University of Tokyo to perform image-based single-cell platelet aggregates analysis that is associated with COVID-19. A pilot paper is published in Nature Communications. We also collaborate with physicians and researchers from School of Medicine University of Virginia to develop a UNet-based segmentation framework for brain hemorrhage CT scans.
Left panel: Image-based facial weakness detection; middle panel: Video-based facial weakness detection; right panel: Abnormal eye movement assessment.
We develop an automated video-based system that can detect common neurological deficits in stroke such as facial weakness using computer vision and biomedical imaging analysis. One of our studies regarding facial weakness detection showed that the system was able to achieve the equal performance to paramedics. Significance of this project is that it provides a proof-of-concept study, showing that such video-based facial weakness diagnosis system could offer clinical assistance to non-neurologists such as paramedics to increase the coverage of neurological prehospital care. Besides, we also perform webcam-based abnormal eye movement quantification and investigate to detect limb drift based human action analysis. This interdisciplinary research is conducted in a close and strong collaboration with physicians and researchers at School of Medicine and University of Virginia Medical Center.
Left panel: Doppler radar based systems for sleep monitoring and vital sign estimation; middle panel: GaitTracker project; right panel: SmartInsole project.
My past research is in the realm of AI-enabled unobtrusive sensing and wearable devices for healthcare, including sensor fusion, microcontroller programming, biomedical signal analysis, and software prototyping. I built several cyber-physical systems that are capable of performing sleep monitoring, cardiac motion characterization for authentication, air quality monitoring, and gait parameters estimation.
Please check my Google Scholar for the latest publications.
Journal papers