EECS/VISE Spring Seminar: Rolando Estrada, Ph.D.
Speaker: Rolando Estrada, PhD,
Research Scientist, Teledyne Scientific and Imaging
Date: Monday, February 27, 2017
Time: 4:15pm
Place: FGH, Room 136
Title: Automating Retinal Vessel Analysis: Topology Estimation and Artery-Vein Classification
Abstract: The retinal vasculature is clinically important, not only for diagnosing ocular diseases, including diabetic retinopathy and glaucoma, but also for monitoring conditions with general cardiovascular manifestation, such as atherosclerosis and high blood pressure. However, manually analyzing retinal vessels is time-consuming, subjective, and error-prone. In this talk, Rolando Estrada will present an automatic, graph-theoretic framework for retinal vessel analysis, with which he will address two fundamental problems: (1) estimating the 3D vascular topology from a single 2D fundus image and (2) automatically classifying retinal arteries and veins. His framework regularizes these inverse problems via a generative tree-growth model and expert, domain-specific features. He will show how to efficiently maximize his model through heuristic search and present state-of-the-art experimental results which confirm the effectiveness of the approach. He will also discuss future research directions, which include: estimating additional vascular properties (e.g., tortuosity or layer location), exploring other imaging domains, and leveraging deep learning techniques.
Speaker Bio: Rolando Estrada holds a Ph.D. in computer science and a M.S. in biomedical engineering from Duke University; he carried out postdoctoral research in ophthalmology at the Duke University Medical Center. His research spans computer vision, medical imaging, machine learning, and neuroscience. Some of his current and recent projects include: automating retinal vessel analysis and classification; developing novel segmentation and registration algorithms for fundus, MRI, CT, and OCT images; investigating EEG-based interventions to improve memory consolidation; and exploring novel deep learning methods based on reinforcement and transfer learning.