2018 CSE Winter school, 2018.01.08-11, Sonofellce
Title & Abstract
Speaker : 예종철 (카이스트)
TALK TITLE : DEEP LEARNING FOR BIOMEDICAL IMAGE RECONSTRUCTION
ABSTRACT :
Recently, deep learning approaches have achieved significant performance improvement over existing iterative reconstruction methods in various biomedical image reconstruction problems. However, it is still unclear why these deep learning architectures work for specific inverse problems. In this talk, we first review the current state-of-the art deep learning image reconstruction algorithms for various imaging modality such as x-ray CT, MRI, optical imaging, PET, ultrasound, etc. Then, we also introduce recent theoretical efforts from signal processing and applied mathematics which tries to link the deep learning approaches to the classical signal processing approach such as compressed sensing, low-rank matrix completion, wavelets, non-local algorithms, etc. The theoretical understanding so far suggests that the success of deep learning is not from a magical power of a black-box, but rather comes from the power of a novel signal representation using non-local basis combined with data-driven local basis, which is indeed a natural extension of classical signal processing theory.
TALK TITLE : DEEP LEARNING FOR BIOMEDICAL IMAGE RECONSTRUCTION
ABSTRACT :
Recently, deep learning approaches have achieved significant performance improvement over existing iterative reconstruction methods in various biomedical image reconstruction problems. However, it is still unclear why these deep learning architectures work for specific inverse problems. In this talk, we first review the current state-of-the art deep learning image reconstruction algorithms for various imaging modality such as x-ray CT, MRI, optical imaging, PET, ultrasound, etc. Then, we also introduce recent theoretical efforts from signal processing and applied mathematics which tries to link the deep learning approaches to the classical signal processing approach such as compressed sensing, low-rank matrix completion, wavelets, non-local algorithms, etc. The theoretical understanding so far suggests that the success of deep learning is not from a magical power of a black-box, but rather comes from the power of a novel signal representation using non-local basis combined with data-driven local basis, which is indeed a natural extension of classical signal processing theory.
Speaker : 최병욱 (연세대학교 방사선의과학연구소장)
TALK TITLE :
ABSTRACT :
TALK TITLE :
ABSTRACT :