• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.6
• /
• pp.2115-2127
• /
• 2021

Although significant progress has been made in synthesizing visually realistic face images by Generative Adversarial Networks (GANs), there still lacks effective approaches to provide fine-grained control over the generation process for semantic facial attribute editing. In this work, we propose a novel cross channel self-attention based generative adversarial network (CCA-GAN), which weights the importance of multiple channels of features and archives pixel-level feature alignment and conversion, to reduce the impact on irrelevant attributes while editing the target attributes. Evaluation results show that CCA-GAN outperforms state-of-the-art models on the CelebA dataset, reducing Fréchet Inception Distance (FID) and Kernel Inception Distance (KID) by 15~28% and 25~100%, respectively. Furthermore, visualization of generated samples confirms the effect of disentanglement of the proposed model.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.6
• /
• pp.285-292
• /
• 2023

With the wide spread of smart farms and the advancements in IoT technology, it is easy to obtain additional data in addition to crop images. Consequently, deep learning-based crop disease diagnosis research utilizing multimodal data has become important. This study proposes a crop disease diagnosis method using multimodal supervised contrastive learning by expanding upon the multimodal self-supervised learning. RandAugment method was used to augment crop image and time series of environment data. These augmented data passed through encoder and projection head for each modality, yielding low-dimensional features. Subsequently, the proposed multimodal supervised contrastive loss helped features from the same class get closer while pushing apart those from different classes. Following this, the pretrained model was fine-tuned for crop disease diagnosis. The visualization of t-SNE result and comparative assessments of crop disease diagnosis performance substantiate that the proposed method has superior performance than multimodal self-supervised learning.

• Proceedings of the KOSOMBE Conference
• /
• v.1997 no.11
• /
• pp.201-205
• /
• 1997

With the recent advance of Web and its associated technologies, information sharing on distribute computing environments has gained a great amount of attention from many researchers in many application areas, such as medicine, engineering, and business. One basic requirement of distributed medical consultation systems is that geographically dispersed, disparate participants are allowed to exchange information readily with each other. Such software also needs to be supported on a broad range of computer platforms to increase the software's accessibility. In this paper, the development of world-wide-web based medical consultation system or radiology imaging is addressed to provide the platform independence and great accessibility. The system supports sharing of 3-dimensional objects. We use VRML (Virtual Reality Modeling Language), which is the de-facto standard in 3-D modeling on the Web. 3-D objects are reconstructed from CT or MRI volume data using a VRML format, which can be viewed and manipulated easily in Web-browsers with a VRML plug-in. A Marching cubes method is used in the transformation of scanned volume data set to polygonal surfaces of VRML. A decimation algorithm is adopted to reduce the number of meshes in the resulting VRML file. 3-D volume data are often very large-sized, and hence loading the data on PC level computers requires a significant reduction of the size of the data, while minimizing the loss of the original shape information. This is also important to decrease network delays. A prototype system has been implemented (http://netopia.snu.ac.kr/-cyber/). and several sessions of experiments are carried out.

• Clinical and Experimental Reproductive Medicine
• /
• v.24 no.3
• /
• pp.335-340
• /
• 1997

Fluorescence in situ hybridization (FISH) techniques allow the enumeration of chromosome abnormalities and from a great potential for many clinical applications. In order to produce quantitative and reproducible results, expensive tools such as a cooled CCD camera and a computer software are required. We have developed a Chromosome Image Processing System (Chips) using FISH that allows the detection and mapping of the genetic aberrations. The aim of our study, therefore, is to evaluate the capabilities of our original system using a black-and-white video camera. As a model system, three repetitive DNA probes (D18Z1, DXZ1, and DYZ3) were hybridized to variety different clinical samples such as human metaphase spreads and interphase nuclei obtained from uncultured peripheral blood lymphocytes, uncultured amniocytes, and germ cells. The visualization of the FISH signals was performed using our system for image acquisition and pseudocoloring. FISH images were obtained by combining images from each of probes and DAPI counterstain captured separately. Using our original system, the aberrations of single or multiple chromosomes in a single hybridization experiment using chromosomes and interphase nuclei from a variety of cell types, including lymphocytes, amniocytes, sperm, and biopsied blastomeres, were enabled to evaluate. There were no differences in the image quality in accordance with FISH method, fluorochrome types, or different clinical samples. Always bright signals were detected using our system. Our system also yielded constant results. Our Chips would permit a level of performance of FISH analysis on metaphase chromosomes and interphase nuclei with unparalleled capabilities. Thus, it would be useful for clinical purposes.

• The Journal of the Korea Contents Association
• /
• v.10 no.6
• /
• pp.1-9
• /
• 2010

Recently, 3D stereoscopic display such as 3D stereoscopic cinemas and 3D stereoscopic TV is getting a lot of interest. In general, a stereo image can be used in 3D stereoscopic display. In other hands, for 3D auto stereoscopic display, the elemental images should be generated through visualization from every camera in a lens array. Since a lens array consists of several cameras, it takes a lot of time to generate the elemental images with respect to 3D virtual space, specially, if a large bio-medical volume data is in the 3D virtual space, it will take more time. In order to improve the problem, in this paper, we construct an octree for a given bio-medical volume data and then propose a method to generate the elemental images through efficient rendering of the Octree data using GPU. Experimental results show that the proposed method can obtain more improvement comparable than conventional one, but the development of more efficient method is required.

• Journal of Science and Technology Studies
• /
• v.11 no.1
• /
• pp.89-121
• /
• 2011

The majority of previous researchers on body management practices including plastic surgery has agreed that there is a strong connection between social demands of plastic surgery and public exposures of beautiful body-images, which this research intends to analyze further. This study, on the one hand, discovers how body-images are produced and consumed through clinical practices of plastic surgery, particularly, surgeon-patient consultation processes based on the researcher's participant observation on a plastic surgery clinic in Korea, and shows how visualization technologies are mobilized to reconstruct not only boundaries of patients' bodies but also those of medical disciplines by viewing plastic surgery practices as knowledge production activities, on the other hand. While revealing that surgeon-patient consultation is the process to transform patient's bodies to "scientific" objects and visualization technologies have been made to help plastic surgeons to make their disciplines "scientific" ones, this article also pays attention to complicated effects of new imaging technology beyond a mere means of "scientification" of plastic surgery.

• The Journal of the Korea Contents Association
• /
• v.2 no.3
• /
• pp.1-15
• /
• 2002

A noel automated brain region extraction method in single channel MR images for visualization and analysis of a human brain is presented. The method generates a volume of brain masks by automatic thresholding using a dual curve fitting technique and by 3D morphological operations. The dual curve fitting can reduce an error in clue fitting to the histogram of MR images. The 3D morphological operations, including erosion, labeling of connected-components, max-feature operation, and dilation, are applied to the cubic volume of masks reconstructed from the thresholded Drain masks. This method can automatically extract a brain region in any displayed type of sequences, including extreme slices, of SPGR, T1-, T2-, and PD-weighted MR image data sets which are not required to contain the entire brain. In the experiments, the algorithm was applied to 20 sets of MR images and showed over 0.97 of similarity index in comparison with manual drawing.

• Progress in Medical Physics
• /
• v.17 no.3
• /
• pp.144-152
• /
• 2006

The telemedicine using independent client-server system via networks can provide high quality normalized services to many hospitals, specifically to local/rural area hospitals. This will eventually lead to a decreased medical cost because the centralized institute can handle big computer hardware systems and complicated software systems efficiently and economically, Customized cancer radiation treatment planning for each patient Is very useful for both a patient and a doctor because it makes possible for the most effective treatment with the least possible dose to patient. Radiation planners know that too small a dose to the tumor can result in recurrence of the cancer, while too large a dose to healthy tissue can cause complications or even death. The best solution is to build an accurate planning simulation system to provide better treatment strategies based on each patient's computerized tomography (CT) image. We are developing a web-based and a network-based customized cancer radiation therapy simulation system consisting of four Important computer codes; a CT managing code for preparing the patients target data from their CT image files, a parallel Monte Carlo high-energy beam code (PMCEPT code) for calculating doses against the target generated from the patient CT image, a parallel linear programming code for optimizing the treatment plan, and scientific data visualization code for efficient pre/post evaluation of the results. The whole softwares will run on a high performance Beowulf PC cluster of about 100-200 CPUs. Efficient management of the hardware and software systems is not an easy task for a hospital. Therefore, we integrated our system into the client-sewer system via network or web and provide high quality normalized services to many hospitals. Seamless communication with doctors is maintained via messenger function of the server-client system.

• Proceedings of the KOSOMBE Conference
• /
• v.1997 no.11
• /
• pp.357-360
• /
• 1997

For medical students and doctors, knowledge of the 3-dimensional (3D) structure of the heart is very important in diagnosis and treatment of the heart diseases. 2-dimensional (2D) tools (e.g. anatomy book) or classical 3D tools (e.g. plastic model) are not sufficient or understanding the complex structures of the heart. Moreover, it is not always guaranteed to dissect the heart of cadaver when it is necessary. To overcome this problem, virtual dissection systems of the heart have been developed. But these systems are not satisfactory since they are made of radiographs; they are not true 3D images; they can not be used to dissect freely; or they can only be operated on the workstation. It is also necessary to make the dissection systems incorporating the various races and tribes because of the organ's difference according to race and tribe. This study was intended to make the 3D image of the heart from a Korean cadaver, and to establish a virtual dissection system of the heart with a personal computer. The procedures or manufacturing this system were as follows. 1. The heart from a Korean adult cadaver was embedded with gelatin solution, and serially cross-sectioned at 1mm-thickness on a meat slicer. Pictures or 153 cross-sectioned specimens were inputted into the computer using a digital camera ($756{\times}504$ resolution, true color). 2. The alignment system was established by means of the language of IDL, and applied to align 2D images of the heart. In each of 2D images, closed curves lining clean and dirty blood pathways were drawn manually on the CorelDRAW program. 3. Using the language of IDL, the 3D image and the virtual dissection system of the heart were constructed. The virtual dissection system of the heart allowed or ree rotation, any-directional sectioning, and selected visualization of the heart's structure. This system is expected to become more advanced, and to be used widely through Internet or CD-title as an educational tool for medical students and doctors.

• Journal of Korea Multimedia Society
• /
• v.13 no.3
• /
• pp.430-437
• /
• 2010

It is essential for living donor liver transplantation that surgeon must understand the hepatic vessel structure to improve the success rate of operation. In this paper, we extract the liver boundary without other surrounding structures such as heart, stomach, and spleen using the contrast enhanced MDCT liver image sequence. After that, we extract the major hepatic veins (left, middle, right hepatic vein) with morphological filter after review the basic structure of hepatic vessel which reside in segmented liver image region. The purpose of this study is provide the overall status of transplantation operation with size estimation of resection part which is dissected along with the middle hepatic vein. The method of liver extraction is as follows: firstly, we get rid of background and muscle layer with gray level distribution ratio from sampling process. secondly, the coincident images match with unit mesh image are unified with resulted image using the corse coordinate of liver and body. thirdly, we extract the final liver image after expanding and region filling. Using the segmented liver images, we extract the hepatic vessels with morphological filter and reversed the major hepatic vessels only with a results of ascending order of vessel size. The 3D reconstructed views of hepatic vessel are generated after applying the interpolation to provide the smooth view. These 3D view are used to estimate the dissection line after identify the middle hepatic vein. Finally, the volume of resection region is calculated and we can identify the possibility of successful transplantation operation.