I am a Ph.D student of UCLA Computer Science , doing research at UCLA under Prof. Majid Sarrafzadeh . I received my Master's in Computer Science from UCLA under the joint supervision of Prof. Demetri Terzopoulos and Prof. Petros Faloutsos ( York University, Toronto ) . I earned my Bachelor's/Master's in Computer & Electrical Engineering at National Technical University of Athens,Greece. My PhD work is focused on Wireless Health, Machine Learning, computer vision and sensor fusion. I am a member of the UCLA Embedded and Reconfigurable Systems Lab (ER Lab).
It is a major challenge to train surgeons with the motor skills required to perform laparoscopic surgery. To address this issue the Society of American Gastrointestinal and Endoscopic Surgeons (SAGES) adopted the Fundamentals of Laparoscopic Surgery (FLS) as a standardized tool set for certification and assessment of medical trainees. However, the process of certification currently requires the expensive time commitment of expert medical professionals and trainees lack the tools to objectively assess their own skill level. We tackle these problems by introducing a computer vision system that can fully analyze and track the accredited FLS Peg Transfer task, producing a meaningful quantification of all the relevant events of the performance. Our solution is fully automated, works with the single camera provided in the FLS Box Trainer equipment, and does not require any modification of the available tools. It provides an automated, objective, and quantitative measurement of a surgeon’s performance. It also allows for remote FLS certification tests for trainees that do not have direct access to an FLS certification center. Importantly, our method is a low-cost solution that does not require any additional equipment or preparation, thus lowering the barrier of adoption of FLS certification.
Sharing medical images is becoming increasing crucial for knowledge discovery in medicine and science. An important issue associated with image sharing is the potential for a breach of patient privacy. Prior work has primarily focused on removing identifying metadata such as name, age or sex from image headers. However, there remains the potential for images being identified directly from the image itself. This is particularly true for images of the head. Techniques have been developed previously that remove the entire facial region from the image. This technique is suboptimal in that the radical defacing removes landmarks that may be useful for correlating with other images, but yet are not important for facial recognition. In this paper we present a framework for identifying and removing features from structural head images. Our methodology focuses on geometric features of the face and thus has the potential to be modality independent.
Efficient cloth simulation is an important problem for interactive applications that involve virtual humans, such as computer games. A common aspect of many methods that have been developed to simulate cloth is a linear system of equations, which is commonly solved using conjugate gradient or multi-grid approaches. In this paper, we introduce to the computer gaming community a recently proposed preconditioner, the incomplete Poisson preconditioner (IPP), for conjugate gradient solvers. We show that IPP performs as well as the current state-of-the-art preconditioners, while being much more amenable to standard thread-level parallelism. We demonstrate our results on an 8-core Mac Pro and a 32-core Emerald Rigde system.