제1회 막수 포럼 (The 1st Last Wednesday Forum)
일시: 2013년 3월 27일, 12:00~13:00
장소: 전산학과(E3-1) 4층 오상수 영상강의실
3 Dimensional Shape Modeling and Analysis for Brain Structures
Presenter: 김재일 (박진아 교수님 연구실)
In recent years, there has been great emphasis on developing the surface-based shape analysis of human brain structures due to their power detecting structural abnormalities caused by clinical factors (i.e., aging and neurodegenerative diseases). Although modeling of individual brain structures has been conducted with various shape representations, it is still challenging in terms of the individual shape reconstruction with anatomical regularity and the comparative quantification of the morphological variations in the brain structures. In this session, we present a novel shape modeling method using the Laplacian surface representation and model-based shape analysis methods to characterize the regional shape variations of brain structures.
4D Dynamic PET Reconstruction using Spatio-Temporal Patch-based Low Rank Regularization
Presenter: 김경상 (예종철 교수님 연구실)
Dynamic positron emission tomography (PET) is widely used to measure changes in the bio-distribution of radiopharmaceuticals within the organs of interest over time. However, as the photon counts for each time frame are limited, conventional reconstruction algorithms such as the ordered subset expectation maximization (OSEM) produce noisy reconstructions. To address this problem, many advanced reconstruction algorithms have been developed using various spatio-temporal regularizations. The main goal is to further extend the results and develop a novel spatio-temporal regularization approach that exploits inherent similarities within and across frames to address the issue. One of the main contributions of this algorithm is to demonstrate that such correlations can be exploited using a low rank constraint of overlapping similarity blocks. Another contribution is that we propose a novel optimization method using the concave-convex procedure (CCCP) by exploiting the Legendre-Fenchel transform. Using simulation results and real in vivo experiments, we confirm that the proposed algorithm can significantly improve image quality and extract high quality kinetic parameters.