• Journal of Korea Multimedia Society
• /
• v.23 no.9
• /
• pp.1129-1138
• /
• 2020

The performance of convolutional deep learning networks is generally determined according to parameters of target dataset, structure of network, convolution kernel, activation function, and optimization algorithm. In this paper, a genetic algorithm is used to select the appropriate deep learning model and parameters for Alzheimer's classification and to compare the learning results with preliminary experiment. We compare and analyze the Alzheimer's disease classification performance of VGG-16, GoogLeNet, and ResNet to select an effective network for detecting AD and MCI. The simulation results show that the network structure is ResNet, the activation function is ReLU, the optimization algorithm is Adam, and the convolution kernel has a 3-dilated convolution filter for the accuracy of dementia medical images.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2013.06a
• /
• pp.274-277
• /
• 2013

CT(Computed Tomography)영상에서 선량과 화질은 중요한 요소이다. 선량은 환자에게 직접적으로 악영향을 끼치는 요소이며, 화질은 환자의 병변을 판단하는데 매우 중요하게 작용한다. 반복적 재구성 알고리즘을 이용하면 저선량 영상에서도 고화질의 영상을 얻을 수 있는지 FBP와 정량적, 정성적으로 비교하였다. 촬영 프로토콜은 관전압 80, 100, 120kVp에서 관전류를 동일하게 200mA로 촬영하여 획득하였으며, 정량적 평가를 위해 SD(Standard Deviation), SNR(Signal to Noise Ratio), MTF(Modulation Transfer Function)를 측정하여 분석하였다. 선량은 80kVp일 때 가장 낮았으며, 120kVp일 때 가장 높았다. 80kVp의 영상을 Toshiba 사(社)의 AIDR 3D(Adaptive Iterative Reduction integrated into $^{SURE}Exposure$)로 재구성하고, 120kVp의 영상에 FBP로 재구성한 다음 정량적 비교를 한 결과 AIDR 3D를 적용한 영상의 SD가 낮게 나왔으며, SNR이 높게 나타났고, MTF 곡선은 유사하게 나타났다. 그리고 FWHM(Full Width at Half Maximum) 값의 오차가 거의 없었다. 결론적으로 AIDR 3D는 저선량에서도 높은 화질을 나타냄을 확인하였다.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.21 no.6
• /
• pp.767-774
• /
• 2011

In recent years, wind turbine system has been considered as the most efficient renewable energy source. Wind turbine system is a complex system which is composed of blade, generator and inverter systems. Recently, lots of researches on fault detection and diagnosis of wind turbine system have been done. Most of them are related with the fault diagnosis of mechanical elements using bivration signal. In this work, a new type of inverter fault detection and diagnstic algorithm is proposed. Furthermore, extensive simulation studies and practical experiments are carried out to verify the proposed algorithm.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.10C
• /
• pp.1402-1413
• /
• 2004

In non-rigid image registration, the Jacobian determinant of the estimated deformation should be positive everywhere since physical deformations are always invertible. We propose a constrained optimization technique at ensures the positiveness of Jacobian determinant for cubic B-spline based deformation. We derived sufficient conditions for positive Jacobian determinant by bounding the differences of consecutive coefficients. The parameter set that satisfies the conditions is convex; it is the intersection of simple half spaces. We solve the optimization problem using a gradient projection method with Dykstra's cyclic projection algorithm. Analytical results, simulations and experimental results with inhale/exhale CT images with comparison to other methods are presented.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.05a
• /
• pp.393-396
• /
• 2021

폐색전증(Pulmonary Embolism : PE)은 사망률이 높은 질환중 하나이다. 초기에 정확한 진단 및 치료를 시행하는 경우 사망과 재발의 경우가 흔치 않으나 지연 진단이 일어나는 경우 사망률은 높기 때문에 보다 정확하고 빠른 검사법이 필요하다. 본 연구에서는 폐색전증 환자의 흉부 CT 데이터를 가지고 딥러닝 알고리즘을 생성하여 소프트웨어를 구현하고자 한다. 개발된 딥러닝 기반의 폐색전증 진단 보조 소프트웨어는 CT 촬영 후 즉각적인 질환 예측으로 판독 시간의 단축과 효율성을 제공할 것으로 기대된다.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.15 no.2
• /
• pp.83-87
• /
• 2011

Purpose: Recently in diagnostics area PET/CT is using a variety of areas including oncology, as well as in cardiology, neurology, etc. While increasing in the importance of PET/CT, there are various researches in the image quality related to reconstruction method. We compared and tested Iterative 2D Reconstruction Method with True X Reconstruction method by Siemens through phantom experiment, so we can see increasing of clinical usefulness of PET/CT. Materials and Methods: We measured contrast ratio and FWHM due to evaluating images on dose and experiment using Biograph 40 True Point PET/CT (Siemens, Germany). Getting a result of contrast ratio and FWHM, we used NEMA IEC PET body phantom (Data Spectrum Corp.) and capillary tube. We used the current TrueX and the previous Iterative 2D algorithm for all images which have 10 minutes long. Also, a clinical suitability of parameter for Iterative 2D and a recommended parameter by Siemens for True X are applied to the experiment. Results: We tested FWHM using capillary tube. As a result, TrueX was less than Iterative 2D. Also, the differences of FWHM get bigger in low dose. On the other hand, we tested contrasts ratio using NEMA IEC PET body phantom. As a result, TrueX was better aspect than Iterative 2D. However, there was no difference in dose. Conclusion: In this experiment, TrueX get higher results of contrast ratio and spatial resolution than Itertive 2D through experiment. Also, in the reconstruction result through TrueX, TrueX had better aspect of resolution than Iterative 2D in low dose. However, contrast ratio had no specific difference. In other words, TrueX reconstruction method in PET/CT had higher clinical value in use because TrueX can reduce exposure of patient and had a better quality of screen.

• Progress in Medical Physics
• /
• v.26 no.1
• /
• pp.28-35
• /
• 2015

In this study, we investigated the effects of an iterative reconstruction algorithm and an automatic exposure control (AEC) technique on image quality and radiation dose through phantom experiments with coronary computed tomography (CT) angiography protocols. We scanned the AAPM CT performance phantom using 320 multi-detector-row CT. At the tube voltages of 80, 100, and 120 kVp, the scanning was repeated with two settings of the AEC technique, i.e., with the target standard deviations (SD) values of 33 (the higher tube current) and 44 (the lower tube current). The scanned projection data were reconstructed also in two ways, with the filtered back projection (FBP) and with the iterative reconstruction technique (AIDR-3D). The image quality was evaluated quantitatively with the noise standard deviation, modulation transfer function, and the contrast to noise ratio (CNR). More specifically, we analyzed the influences of selection of a tube voltage and a reconstruction algorithm on tube current modulation and consequently on radiation dose. Reduction of image noise by the iterative reconstruction algorithm compared with the FBP was revealed eminently, especially with the lower tube current protocols, i.e., it was decreased by 46% and 38%, when the AEC was established with the lower dose (the target SD=44) and the higher dose (the target SD=33), respectively. As a side effect of iterative reconstruction, the spatial resolution was decreased by a degree that could not mar the remarkable gains in terms of noise reduction. Consequently, if coronary CT angiogprahy is scanned and reconstructed using both the automatic exposure control and iterative reconstruction techniques, it is anticipated that, in comparison with a conventional acquisition method, image noise can be reduced significantly with slight decrease in spatial resolution, implying clinical advantages of radiation dose reduction, still being faithful to the ALARA principle.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.29 no.1
• /
• pp.36-42
• /
• 2009

We have developed software for converting the volumetric voxel data obtained from X-ray computed tomography(CT) into computer-aided design(CAD) data. The developed software can used for non-destructive testing and evaluation, reverse engineering, and rapid prototyping, etc. The main algorithms employed in the software are image reconstruction, volume rendering, segmentation, and mesh data generation. The feasibility of the developed software is demonstrated with the CT data of human maxilla and mandible bones.

• Journal of KIISE:Computing Practices and Letters
• /
• v.7 no.5
• /
• pp.451-457
• /
• 2001

This paper describes automated methods for the detection of lung nodules and their volume calculation on CT scans. Gray-level threshold methods were used to segment the thorax from the background and then the lung parenchymes from the thoracic wall and mediastinum. A scanning-ball algorithm was applied to more accurately delineate the lung boundaries, thereby incorporating peripheral nodules contiguous to pleural surface within the segmented lung parenchymes. The lesions which have the high gray value were extracted from the segmented lung parenchymes. The selected lesions include nodules, blood vessels and partial volume effects. The discriminating features such as size, solid-shape, average, standard deviation and correlation coefficient of selected lesions were used to distinguish true nodules from pseudo-lesions. Volume and circularity calculation were performed for each identified nodules. The identified nodules were sorted in descending order of the volume. These method were applied to 621 image slices of 19 cases. The sensitivity was 95% and there was no false-positive result.

• Proceedings of the KIEE Conference
• /
• 2007.07a
• /
• pp.44-45
• /
• 2007

This paper deals with error compensation in current transformers. Since the exciting current can be considered as the main error source, its evaluation can allow the compensation of its detrimental effects to be obtained. The exciting current required by the transformer in every king of steady state operation can be determined by simply acquiring the secondary current, provided that the examined CT has been preliminarily identified. This paper also proposed a new approach to the model of the exciting branch. The exciting branch can be divided into a non-linear core loss resistor, and a non-linear magnetizing inductor whose flux and current characteristic is not the same as the characteristic shown by the joined tips of the first quadrant of a family of hysteresis loops. The performance of the proposed algorithm was validated under various conditions using EMTP generated data. Test result show, in all cases an improvement in primary current reproduction accuracy, compared with that achieved using CT's ratio.