• 한국멀티미디어학회논문지
• /
• 제25권5호
• /
• pp.757-768
• /
• 2022

In this paper, we proposed a system that visualizes a hologram device in 3D by utilizing the CT image segmentation function based on artificial intelligence deep learning. The input axial CT medical image is converted into Sagittal and Coronal, and the input image and the converted image are divided into 3D volumes using ResUNet, a deep learning model. In addition, the volume is created by segmenting the tumor region in the segmented liver image. Each result is integrated into one 3D volume, displayed in a medical image viewer, and converted into a video. When the converted video is transmitted to the hologram device and output from the device, a 3D image with a sense of space can be checked. As for the performance of the deep learning model, in Axial, the basic input image, DSC showed 95.0% performance in liver region segmentation and 67.5% in liver tumor region segmentation. If the system is applied to a real-world care environment, additional physical contact is not required, making it safer for patients to explain changes before and after surgery more easily. In addition, it will provide medical staff with information on liver and liver tumors necessary for treatment or surgery in a three-dimensional manner, and help patients manage them after surgery by comparing and observing the liver before and after liver resection.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2010년도 춘계학술대회 논문집
• /
• pp.71-72
• /
• 2010

• 대한임베디드공학회논문지
• /
• 제16권5호
• /
• pp.215-223
• /
• 2021

3D data can be categorized into two parts : Euclidean data and non-Euclidean data. In general, 3D data exists in the form of non-Euclidean data. Due to irregularities in non-Euclidean data such as mesh and point cloud, early 3D deep learning studies transformed these data into regular forms of Euclidean data to utilize them. This approach, however, cannot use memory efficiently and causes loses of essential information on objects. Thus, various approaches that can directly apply deep learning architecture to non-Euclidean 3D data have emerged. In this survey, we introduce various deep learning methods for mesh and point cloud data. After analyzing the operating principles of these methods designed for irregular data, we compare the performance of existing methods for shape classification and segmentation tasks.

• Journal of Biosystems Engineering
• /
• 제27권3호
• /
• pp.227-234
• /
• 2002

This study was conducted to nondestructively estimate the volumetric information of peach and pit and to visualize the 3D information of internal structure from magnetic resonance imaging(MRI) data. Bruker Biospec 7T spectrometer operating at a proton reosonant frequency of 300 MHz was used for acquisition of MRI data of peach. Image processing algorithms and visualization techniques were implemented by using MATLAB (Mathworks) and Visualization Toolkit(Kitware), respectively. Thresholding algorithm and Kohonen's self organizing map(SOM) were applied to MRI data fur region segmentation. Volumetric information were estimated from segemented images and compared to the actual measurements. The average prediction errors of peach and pit volumes were 4.5%, 26.1%, respectively for the thresholding algorithm. and were 2.1%, 19.9%. respectively for the SOM. Although we couldn't get the statistically meaningful results with the limited number of samples, the average prediction errors were lower when the region segmentation was done by SOM rather than thresholding. The 3D visualization techniques such as isosurface construction and volume rendering were successfully implemented, by which we could nondestructively obtain the useful information of internal structures of peach.

• Journal of International Society for Simulation Surgery
• /
• 제2권1호
• /
• pp.17-25
• /
• 2015

Purpose In this paper, we propose a robust 3D vessel tracking algorithm by utilizing an active contour model and unscented Kalman filter which are the two representative algorithms on segmentation and tracking. Materials and Methods The proposed algorithm firstly accepts user input to produce an initial estimate of vessel boundary segmentation. On each Computed Tomography Angiography (CTA) slice, the active contour is applied to segment the vessel boundary. After that, the estimation process of the unscented Kalman filter is applied to track the vessel boundary of the current slice to estimate the inter-slice vessel position translation and shape deformation. Finally both active contour and unscented Kalman filter are inter-operated for vessel segmentation of the next slice. Results The arbitrarily shaped blood vessel boundary on each slice is segmented by using the active contour model, and the Kalman filter is employed to track the translation and shape deformation between CTA slices. The proposed algorithm is applied to the 3D visualization of chest CTA images using graphics hardware. Conclusion Through this algorithm, more opportunities, giving quick and brief diagnosis, could be provided for the radiologist before detailed diagnosis using 2D CTA slices, Also, for the surgeon, the algorithm could be used for surgical planning, simulation, navigation and rehearsal, and is expected to be applied to highly valuable applications for more accurate 3D vessel tracking and rendering.

• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제4권11호
• /
• pp.529-536
• /
• 2015

비사실적 렌더링(NPR; Non-Photorealistic Rendering)은 2차원 영상과 3차원 모델을 대상으로 하는 방법이 다르며 각각의 대상에 NPR을 적용하여 두 콘텐츠를 혼합하면 이질감이 나타나는 문제점이 있다. 본 논문에서는 3차원 객체와 영상에 있어서 각각의 대상에 카툰 및 스케치와 같은 비사실적 효과를 적용하여 조화롭게 혼합하는 기법을 제시한다. 제안 기법은 2차원 영상의 데이터를 분석하여 컬러 분포 특징을 얻고 이를 이용하여 실사 영상이나 3D 객체의 컬러 수를 줄인다. 단순화된 컬러맵과 윤곽선 에지 데이터로부터 비사실적 렌더링을 실시한다. 컬러맵 정보의 추출 및 적용 과정에서 자연스러운 장면 연출을 위해서 영상분할 과정이 필요하다. 그러나 영상분할 기법은 많은 연산을 필요로 한다. 특히 크기가 큰 입력에 대해서는 비사실적 렌더링에 많은 시간이 소요된다. 처리 시간이 많은 영상분할의 고속화를 위하여 GPU(Graphics Processing Unit)를 이용한 병렬 컴퓨팅을 할 수 있는 GPGPU(General-Purpose GPU)를 사용한다. GPGPU의 사용으로 알고리즘의 수행속도를 크게 개선하였다. 또한 영상분할 후 단순화된 컬러를 추출하여 일련의 컬러맵을 생성한 뒤 3D 객체에 NPR을 적용할 때 추출해낸 컬러맵을 적용하여 2차원 영상과 3차원 객채 간의 이질감을 줄이고 조화롭게 하였다.

• 터널과지하공간
• /
• 제33권6호
• /
• pp.508-518
• /
• 2023

이 논문은 LiDAR 스캔 또는 사진측량 기술에 의해 재구성된 3D 디지털 모델을 기반으로 터널 벽면의 불연속면을 자동으로 매핑하는 새로운 접근 방식을 제안한다. 본 제안에서는 U-Net이라 불리는 딥러닝 시맨틱 영역분할 모델을 사용하며, 터널 막장면의 3D 지형 모델에서 불연속면 영역을 식별해 낸다. 제안된 딥러닝 모델은 투영된 RGB 이미지, 면의 깊이 이미지 및 국부적인 면의 표면 속성 이미지(즉, 법선 벡터 및 곡률 이미지)를 포함한 다양한 정보를 종합 학습하여 기본 3차원 이미지에서 불연속면 영역을 효과적으로 분할한다. 이후 영역분할 결과는 면의 깊이 맵과 투영 행렬을 사용하여 3D 모델로 다시 투영시키고, 3D 공간 내에서 불연속면의 위치 및 범위를 정확하게 표현한다. 영역분할 모델의 성능은 영역 분할된 결과를 해당 지면 실측 값과 비교함으로써 평가하였으며, IoU(intersection-over-union) 값이 약 0.8 정도로 나타나 영역분할 결과의 높은 정확성을 확인하였다. 여전히 학습데이터가 제한적 이었음에도 불구하고, 제안 기법은 3D 모델의 점군 데이터를 불연속면의 유사군으로 그룹화하기 위해 전 막장면의 법선 벡터와 클러스터링과 같은 비지도 학습기반 알고리즘에만 의존하던 기존 접근 방식의 한계의 극복 가능성을 보여주었다.

• 한국의학물리학회지:의학물리
• /
• 제32권4호
• /
• pp.172-178
• /
• 2021

Purpose: This study aimed to develop a deep learning architecture combining two task models to generate synthetic computed tomography (sCT) images from low-tesla magnetic resonance (MR) images to improve metallic marker visibility. Methods: Twenty-three patients with cervical cancer treated with intracavitary radiotherapy (ICR) were retrospectively enrolled, and images were acquired using both a computed tomography (CT) scanner and a low-tesla MR machine. The CT images were aligned to the corresponding MR images using a deformable registration, and the metallic dummy source markers were delineated using threshold-based segmentation followed by manual modification. The deformed CT (dCT), MR, and segmentation mask pairs were used for training and testing. The sCT generation model has a cascaded three-dimensional (3D) U-Net-based architecture that converts MR images to CT images and segments the metallic marker. The performance of the model was evaluated with intensity-based comparison metrics. Results: The proposed model with segmentation loss outperformed the 3D U-Net in terms of errors between the sCT and dCT. The structural similarity score difference was not significant. Conclusions: Our study shows the two-task-based deep learning models for generating the sCT images using low-tesla MR images for 3D ICR. This approach will be useful to the MR-only workflow in high-dose-rate brachytherapy.

• 한국콘텐츠학회논문지
• /
• 제9권6호
• /
• pp.266-274
• /
• 2009

복부대동맥류는 주로 65-75세의 중년이후 남성과 흡연자에서 주로 발생한다. 가장 중요한 증세는 대동맥 파열로서 생명에 치명적이며, 혈관벽이 헐고 약해지고 파열되어 많은 양의 혈액이 복강 내로 쏟아지는 것을 의미한다. 복부대동맥박리를 치료하기 위해서는 3차원 영상 정보가 필요하고, 수술시 임상의사에게 많은 도움이 된다. 3차원 정보는 MDCT로부터 계산되고 3차원 모델은 2차원 CT영상의 분할로 계산된 좌표로부터 재구성된다. 따라서 3차원 영상의 질은 2차원 영상의 분할알고리듬에 의존적이다. 본 연구에서는 목적장기만을 모델링하기 위해서 가변형 능동모델을 제안한다. 가변형 능동모델은 외부힘에 의해서 에너지가 최소화되는 수렴하는 모델이다. 외부힘은 GVF로 불리며, 그레이레벨 또는 영상으로 부터의 이 진경계지도의 구배가 확산되는 것을 계산한다. 실험결과 복부대동맥박리에 적용해서 3차원 표면재구성을 성공했으며, 분할알고리듬의 특성으로 시각적 및 정량적인 평가도 성공했다.

• 한국의류학회지
• /
• 제33권1호
• /
• pp.68-79
• /
• 2009

It is inevitable to have cutting line to get the 2D pattern from 3D body surface. In this paper the efficiency of curvature plot as a cutting line in the process of flattening 3D surface was investigated. As reference, basic clothing construction line was adopted to divide the 3D surface into small blocks to make the flattening process easy. Female dummy as well as human body were scanned and surface of the upper body was segmented using curvature plot and basic constructing line. 2D tight-fit pattern was developed using three software, the RapidForm 2004, 2C-AN and Yuka CAD. Gap between clothes and body, and the clothing pressure on the body was observed to determine the fit of the clothes. As results, clothes constructed with blocks divided by curvature plot displayed a similar level of tight fit as compared with those by basic construction line. It was found that curvature plot is effective method as a segmentation of the 3D surface even for the actual body which does not have any previous reference line. It is expected that application of curvature plot will be expanded in 3D apparel technology.