http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Photon-counting spectral CT 영상에서의 Water cylinder phantom을 이용한 에너지 왜곡 보정
김동혁(Dong-hyeok Kim),백종덕(Jong-duk Baek) 대한전자공학회 2018 대한전자공학회 학술대회 Vol.2018 No.6
Photon-counting detector (PCD) is one of the leading candidates for the next generation X-ray detector with low dose, material decomposition, and high SNR characteristics. PCD has a nonlinear pixel response and deviation for the incident X-ray energies as well as the pulse pile-up phenomenon with high flux. Therefore, accurate energy calibration is an important factor in overall image quality. In this study, we confirmed the response characteristics of the PCD and applied a flat field correction with a water cylinder phantom to reduce the nonlinear energy response of detector pixels. Our results showed clear reduction of ring artifacts caused by energy distortion regardless of energy bins. And we could confirm the linear attenuation coefficient of materials according to energy increase in the reconstructed image.
RED-CNN에 기반한 CT 영상 픽셀 크기에 따른 denoising 성능평가
김성준(Seong-Jun Kim),김병준(Byeong-Joon Kim),백종덕(Jong-Duk Baek) 대한전자공학회 2019 대한전자공학회 학술대회 Vol.2019 No.11
In X-ray CT studies, reducing the noise in CT images is an increasingly important problem. Many studies suggested and focused on how to apply processes of deep learning, such as network types or loss functions. However, we thought there was an efficient way of preprocessing for better denoising performance of a deep learning method. In this study, we used fan beam geometry to generate noise in CT images, and changed the pixel size of the images. We expected that the smaller pixel size of the images brings out the better effect of denoising by RED-CNN. This study showed how CT images quality changed by pixel size of the images.
Convolutional Neural Network Architecture 에 따른 Cone beam artifact 제거 성능 비교
오준호(Jun-Ho Oh),김병준(Byeong-Joon Kim),백종덕(Jong-Duk Baek) 대한전자공학회 2018 대한전자공학회 학술대회 Vol.2018 No.11
CBCT(Cone beam CT) is widely used in diagnosis and treatment planning of implant dentistry, orthopedics, and interventional radiology. However, the reconstructed images by CBCT geometry generate cone beam artifacts, which would disturb lesion detectability and degrade diagnostic accuracy. In this work, we present convolutional neural network(CNN) based cone beam artifacts correction method, which is computationally efficient and achieve better performance in artifact correction. We compared the performance of the cone beam artifacts reduction via U-Net and ResNet models, which are trained with simulated CBCT images. Our result showed that U-Net performs better than ResNet in cone beam artifact reduction.
RED-CNN에 기반한 X-ray dose에 따른 CT 영상에서의 denoising 평가
김성준(Seong-Jun Kim),김병준(Byeong-Joon Kim),백종덕(Jong-Duk Baek) 대한전자공학회 2019 대한전자공학회 학술대회 Vol.2019 No.6
Reducing the radiation dose to the patients has been an important issue in X-ray CT research. Although several iterative based techniques have been proposed, they are time-consuming, and sometime need to find a proper optimization parameter for the given tasks. Recently, convolutional neural network has shown a big success in denoising of natural images, and its application on low dose CT denoising using RED-CNN (Residual Encode-Decode Convolutional Neural Network) shows a promising result. In this work, we evaluate the performance of the RED-CNN in low dose CT denoising for different dose levels, and how its image quality changes after passing through the RED-CNN.