http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
A Distance-Driven Deconvolution Method for CT Image-Resolution Improvement
Seokmin Han,Kihwan Choi,Sang Wook Yoo,Jonghyon Yi 한국물리학회 2016 THE JOURNAL OF THE KOREAN PHYSICAL SOCIETY Vol.69 No.12
The purpose of this research is to achieve high spatial resolution in CT (computed tomography) images without hardware modification. The main idea is to consider geometry optics model, which can provide the approximate blurring PSF (point spread function) kernel, which varies according to the distance from the X-ray tube to each point. The FOV (field of view) is divided into several band regions based on the distance from the X-ray source, and each region is deconvolved with a different deconvolution kernel. As the number of subbands increases, the overshoot of the MTF (modulation transfer function) curve increases first. After that, the overshoot begins to decrease while still showing a larger MTF than the normal FBP (filtered backprojection). The case of five subbands seems to show balanced performance between MTF boost and overshoot minimization. It can be seen that, as the number of subbands increases, the noise (STD) can be seen to show a tendency to decrease. The results shows that spatial resolution in CT images can be improved without using high-resolution detectors or focal spot wobbling. The proposed algorithm shows promising results in improving spatial resolution while avoiding excessive noise boost.
Seokmin Han,Suchul Lee,Jun-Rak Lee 한국인터넷방송통신학회 2019 Journal of Advanced Smart Convergence Vol.8 No.1
In this research, a practical deep learning framework to differentiate the lesions and nodules in breast acquired with ultrasound imaging has been proposed. 7408 ultrasound breast images of 5151 patient cases were collected. All cases were biopsy proven and lesions were semi-automatically segmented. To compensate for the shift caused in the segmentation, the boundaries of each lesion were drawn using Fully Convolutional Networks(FCN) segmentation method based on the radiologist’s specified point. The data set consists of 4254 benign and 3154 malignant lesions. In 7408 ultrasound breast images, the number of training images is 6579, and the number of test images is 829. The margin between the boundary of each lesion and the boundary of the image itself varied for training image augmentation. The training images were augmented by varying the margin between the boundary of each lesion and the boundary of the image itself. The images were processed through histogram equalization, image cropping, and margin augmentation. The networks trained on the data with augmentation and the data without augmentation all had AUC over 0.95. The network exhibited about 90% accuracy, 0.86 sensitivity and 0.95 specificity. Although the proposed framework still requires to point to the location of the target ROI with the help of radiologists, the result of the suggested framework showed promising results. It supports human radiologist to give successful performance and helps to create a fluent diagnostic workflow that meets the fundamental purpose of CADx.
Han, Seokmin,Lee, Suchul,Lee, Jun-Rak The Institute of Internet 2019 Journal of Advanced Smart Convergence Vol.8 No.1
In this research, a practical deep learning framework to differentiate the lesions and nodules in breast acquired with ultrasound imaging has been proposed. 7408 ultrasound breast images of 5151 patient cases were collected. All cases were biopsy proven and lesions were semi-automatically segmented. To compensate for the shift caused in the segmentation, the boundaries of each lesion were drawn using Fully Convolutional Networks(FCN) segmentation method based on the radiologist's specified point. The data set consists of 4254 benign and 3154 malignant lesions. In 7408 ultrasound breast images, the number of training images is 6579, and the number of test images is 829. The margin between the boundary of each lesion and the boundary of the image itself varied for training image augmentation. The training images were augmented by varying the margin between the boundary of each lesion and the boundary of the image itself. The images were processed through histogram equalization, image cropping, and margin augmentation. The networks trained on the data with augmentation and the data without augmentation all had AUC over 0.95. The network exhibited about 90% accuracy, 0.86 sensitivity and 0.95 specificity. Although the proposed framework still requires to point to the location of the target ROI with the help of radiologists, the result of the suggested framework showed promising results. It supports human radiologist to give successful performance and helps to create a fluent diagnostic workflow that meets the fundamental purpose of CADx.
Seokmin Han,Kihwan Choi,Sang Wook Yoo 한국물리학회 2016 THE JOURNAL OF THE KOREAN PHYSICAL SOCIETY Vol.68 No.6
The objective of this research is to suggest an image synthesis method to simulate the change in the incident X-ray spectrum, based on knowledge of the tissue thickness. The image of a reference phantom was acquired before acquiring the image of an object under the same condition. With knowledge of the thickness of an input pixel, the mapping region could be limited to a specific thickness region in the reference phantom. Taking advantage of the relation between the reference phantom and the imaging object mapped to a limited region, we could synthesize an image acquired at a different tube current and tube voltage. The noise difference was compensated for by the addition of zero-mean Gaussian noise, and RMSE(root mean-square error) between the synthesized image and the actual image was calculated. The average noise difference between the synthesized image and the actual image was about 4.5% (ranging from 0% to 7.7%). RMSE between those images was calculated to be about 1.4% of the average pixel intensity. The calculated results indicate that the proposed method can provide a synthesized image close to the actual image.
한석민 한국교통대학교 2021 한국교통대학교 논문집 Vol.56 No.-
In Deep Learning method, it is well known that it requires large amount of data to train the deep neural network. What makes the learning process more difficult is that it also requires the annotation of each data to fully train the neural network, which indicates that experts should provide the annotation through time-consuming labeling process. To alleviate the problem of time-consuming labeling process, some methods have been suggested such as weak-supervised method, one-shot learning, self-supervised, suggestive learning, and so on. In this manuscript, the researches that apply the suggestive learning are analyzed and its possible future direction of the research is suggested.
심층학습을 이용한 Railroad defect detection기법 분석 및 현황
한석민 韓國交通大學校 2022 한국교통대학교 논문집 Vol.57 No.-
Train is one of the most popular forms of public transportation. Therefore, unexpected accidents and delays are considered serious for railways, which make maintenance process essential. Currently, the inspection of the rail and the fasteners on the railway track is mainly operated by railway staff. Computer vision based methods are now being employed to detect the defects of rails and inspect the railroad condition, so that the high cost of the inspection by railroad investigation staff and low efficiency could be alleviated. Automated defect detection and segmentation can help investigators find rail defects. In this paper, the researches that applied computer vision based deep learning method to railroad defect detection and inspection have been reviewed, and the current trend and the direction of those researches were discussed.