检索结果-内蒙古大学图书馆

IEEE Symposium on Nuclear Science (NSS/MIC)

作者： S. Y. Kim H. B. Chung J. S. Lee M. J. Kang Interdisciplinary Program in Artificial Intelligence Seoul National University Seoul Republic of Korea Seoul National University Nuclear Medicine Bioengineering and Biomedical Sciences Seoul Republic of Korea Seoul National University Mathematical Sciences Seoul Republic of Korea

ISBN: (数字)9798350388152

ISBN: (纸本)9798350388169

In the medical domain, the scarcity of data leads to class imbalances and fairness issues, significantly impairing deep learning performance and leading to the underrepresentation of minority patient groups. To address these challenges, data augmentation techniques employing Generative Adversarial Networks (GANs) and diffusion models have been extensively explored, with diffusion models particularly noted for their ability to enhance data diversity and quality. Despite their achievements, it remains a challenging task to generate a diverse range of samples from limited datasets and to achieve precise generation of images that align with user *** overcome these limitations, we propose a framework, EDM (Enhancing Quality Diversity of Medical image augmentation under the user-specified complex and various conditions), designed to generate diverse high-quality medical images under the various and complex multimodal conditions, representing age, sex, and diseases. Our framework comprises four steps. First, we propose a metric to quantize the necessity of data augmentation. Second, we prepare data augmentation. We generate medical sketch-image-text triplets using DiffSketcher from the real patients’ dataset. Then we employ these triplets to train ControlNet, enabling the generation of synthetic medical images from sketch data. Third, we generate synthetic medical images. We obtain multiple sketches for each image from a diversity-enhanced DiffSketcher, incorporating similarity loss. Using these sketches, we augment data from the fine-tuned ControlNet from the second step. Lastly, we introduce a metric to measure the diversity of the generated images, ensuring the effectiveness of our augmentation *** preliminary experimental results imply EDM can be the promising solution. Future experiments are planned to construct synthetic datasets of various modalities such as PET and CT for underrepresented patient groups.

关键词： Semiconductor device measurement Microwave integrated circuits Semiconductor detectors Data augmentation Diffusion models Generative adversarial networks Image augmentation

来源：评论

学校读者我要写书评

暂无评论

A Novel Energy Estimation Method for PET Detectors Composed of Multiple Scintillators

A Novel Energy Estimation Method for PET Detectors Composed ...

引用

IEEE Symposium on Nuclear Science (NSS/MIC)

作者： H. S. Shim M. J. Cho J. S. Lee Interdisciplinary Program of Bioengineering Seoul National University Seoul Republic of Korea Integrated Major in Innovative Medical Science Seoul National University Seoul Republic of Korea Department of Nuclear Medicine Seoul National University Seoul Republic of Korea

In this study, we propose a novel technique for estimating energy in PET detectors composed of multiple scintillators. We mixed waveforms from two detectors, each containing BGO, GAGG, or LGSO scintillators, and trained a model using artificial neural networks to distinguish the energy of each waveform. We compared the accuracy of using amplitude and integration values as labels for energy estimation, using R 2 values as a metric. When applying the neural network-based energy estimator to the mixed waveforms of BGO and GAGG detectors, the R 2 values were (0.88, 0.99) for amplitude-based training and (0.95, 0.99) for integration-based training. In BGO and LGSO, the R 2 values were (0.94, 0.99) for amplitude-based training and (0.99, 0.99) for integration-based training. In GAGG and LGSO, the R 2 values were (0.99, 0.98) for amplitude-based training and (0.99, 0.98) for integration-based training. By utilizing this method, we could estimate energy in radiation detectors that combine scintillators with different properties, including phoswich detectors and meta-scintillators.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Unobtrusive Monitoring of ECG-derived Features During Daily Smartphone Use

Unobtrusive Monitoring of ECG-derived Features During Daily ...

引用

Annual International Conference of the IEEE Engineering in Medicine and Biology Society

作者： Sungjun Kwon Seungwoo Kang Youngki Lee Chungkuk Yoo Kwangsuk Park Interdisciplinary Program of Bioengineering Seoul National University Department of Computer Science Korea Advanced Institute of Science and Technology (KAIST) Department of Information Systems at Singapore Management University Department of Biomedical Engineering College of Medicine Seoul National University

ISBN: (纸本)9781424479276

Heart rate variability (HRV) is known to be one of the representative ECG-derived features that are useful for diverse pervasive healthcare applications. The advancement in daily physiological monitoring technology is enabling monitoring of HRV in people's everyday lives. In this study, we evaluate the feasibility of measuring ECG-derived features such as HRV, only using the smartphone-integrated ECG sensors system named Sinabro. We conducted the evaluation with 13 subjects in five predetermined smartphone use cases. The result shows the potential that the smartphone-based sensing system can support daily monitoring of ECG-derived features;The average errors of HRV over all participants ranged from 1.65% to 5.83% (SD: 2.54～10.87) for five use cases. Also, all of individual HRV parameters showed less than 5% of average errors for the three reliable cases.

关键词： Power system stability Circuit stability Stability criteria Transient analysis Steady-state

来源：评论

学校读者我要写书评

暂无评论

Accelerated Direct PET Attenuation and Scatter Corrections Using Generation-Prior Diffusion Model (GPDM)

Accelerated Direct PET Attenuation and Scatter Corrections U...

引用

IEEE Symposium on Nuclear Science (NSS/MIC)

作者： M.J. Cho H.S. Shim J.S. Lee Interdisciplinary Program of Bioengineering Seoul National University Seoul Republic of Korea Integrated Major in Innovative Medical Science Seoul National University Seoul Republic of Korea Department of Nuclear Medicine Seoul National University Seoul Republic of Korea

ISBN: (数字)9798350388152

ISBN: (纸本)9798350388169

Attenuation and scatter corrections play pivotal roles in PET image processing for accurate quantification. While CT-based corrections are conventional, they encounter limitations such as respiratory motion and metal artifacts. Recent advances in generative deep neural networks offer promising alternatives by directly estimating attenuation and scatter-corrected (ASC) PET from non-attenuation and non-scatter-corrected (NASC) PET. In GAN-based methods, the generator typically learns to map input data to the desired output (e.g. NASC PET to ASC PET) in one step. While this approach can be effective in generating realistic images, it often encounters significant challenges, such as unstable training and mode collapse. On the other hand, Denosing Diffusion Probabilistic Model (DDPM), a recent innovative solution based on probability theory, has emerged to address these issues, and gradually refines the image over numerous steps (e.g., 1,000 steps). In our previous study, a conditional DDPM (cDDPM) model, where the diffusion process was conditioned with NASC PET to generate ASC PET, outperformed conventional CycleGAN and conditional GAN (cGAN) approaches in the direct PET attenuation and scatter corrections [1]. However, the high computational complexity is one of the main drawbacks of DDPM. The training and inference steps of DDPM can be computationally intensive, requiring significant resources in terms of time, memory, and hardware. To overcome these problems, this study proposes a new acceleration method for the fast learning and sampling process of DDPM. In this method, named Generation-Prior Diffusion Model (GPDM), cDDPM is conditioned with prior variables obtained by applying a U-NET or cGAN to generate ASC PET from NASC PET. GPDM can significantly reduce the sampling steps owing to the high similarity between conditioned images and ASC PET. Consequently, computational complexity can be effectively decreased.

关键词： Training Microwave integrated circuits Semiconductor detectors Image processing Memory management Metals Attenuation

来源：评论

学校读者我要写书评

暂无评论

Application of CT simulation technique for virtual ultra-low-dose trial in CT colonography

Application of CT simulation technique for virtual ultra-low...

引用

4th International Workshop on Computational and Clinical Applications in Abdominal Imaging, Held in Conjunction with the 15th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2012

作者： Kim, Chang Won Kim, Jong Hyo Interdisciplinary Program of Bioengineering Seoul National University College of Engineering San 56-1 Sillim-dong Gwanak-gu Seoul 151-742 Korea Republic of Department of Radiology Seoul National University College of Medicine Seoul 110-744 Korea Republic of Interdisciplinary Program in Radiation Applied Life Science Seoul National University College of Medicine Seoul 110-744 Korea Republic of Institute of Radiation Medicine Seoul National University Medical Research Center 101 Daehangno Jongno-gu Seoul 110-744 Korea Republic of Department of Intelligent Convergence Systems Graduate School of Convergence Science and Technology Seoul National University 864-1 Iui-dong Yeongtong-gu Suwon-si Gyeonggi-do 443-270 Korea Republic of

ISBN: (纸本)9783642336119

A low-dose CT simulation technique is presented which might allow for a virtual ultra-low-dose trial in CT colonography without requiring raw sinogram data. A virtual sinogram is generated by performing the line integral of the CT number-based attenuation value with use of the CT scan parameters available in the DICOM header and in the literature. A separate noise sinogram is generated with use of a noise model, which incorporates the X-ray photon flux depending on the mAs, system electronic noise, and virtual sinogram. A synthetic noise CT image is generated by application of the filtered back projection of the noise sinogram with use of an appropriate filter that depends on the reconstruction kernel of the original CT. Finally, a simulated low-dose CT image is generated by addition of the CT data for the synthetic noise to the original CT data. Clinical CT colonography images with and without fecal tagging were used as simulation input and 50%, 25%, and 12.5% dose images were generated and evaluated. Our results suggest that the proposed CT simulation technique has potential for application in virtual ultra-low-dose trial in CT colonography in which an unlimited number of scan protocols could be performed without repetition of the real CT exposure to the patients. © 2012 Springer-Verlag.

关键词： Computerized tomography

来源：评论

学校读者我要写书评

暂无评论

Comprehensive Study on Cherenkov Radiation for Ultrafast PET Detectors, Considering Refractive Index and Wavelength

Comprehensive Study on Cherenkov Radiation for Ultrafast PET...

引用

IEEE Symposium on Nuclear Science (NSS/MIC)

作者： H.S. Shim H. Kang J. Kim J.S. Lee Y. Jeong Interdisciplinary Program of Bioengineering Seoul National University Seoul Republic of Korea Integrated Major in Innovative Medical Science Seoul National University Seoul Republic of Korea Department of Electrical and Computer Engineering Seoul National University Seoul Republic of Korea

This study aims to investigate the characteristics of Cherenkov radiation with respect to wavelength and refractive index. Refractive index is a function of wavelength, and when it changes, not only does the gain of Cherenkov radiation increase, but also the photon loss in the medium increases. Therefore, we examine how photons of different wavelengths behave under changing refractive index conditions. We adjust the refractive index to maintain a constant emission angle of Cherenkov radiation based on the deceleration rate of electrons moving in the medium. When electrons generating Cherenkov radiation move a total of 200 μm, both long-wavelength and short-wavelength photons are emitted at the initial stage; however, only short-wavelength photons are emitted at the late stage. Although the shorter-wavelength photons arrive first among the photons that reach the detector, the longer-wavelength photons are detected more, forming a significant peak, which is due to the fact that the former experience more absorption-based photon loss in the medium than the latter.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Kidney on a chip

RSC Nanoscience and Nanotechnology

引用

RSC Nanoscience and Nanotechnology 2015年第36期2015-January卷 19-39页

作者： Ha, Laura Jang, Kyung-Jin Suh, Kahp-Yang Interdisciplinary Program of Bioengineering Seoul National University Seoul151-742 Korea Republic of Wyss Institute for Biologically Inspired Engineering Harvard University BostonMA02115 United States School of Mechanical and Aerospace Engineering Seoul National University Seoul151-742 Korea Republic of

来源：评论

学校读者我要写书评

暂无评论

Connecting distant entities with induction through conditional random fields for named entity recognition: Precursor-induced CRF

arXiv

引用

arXiv 2018年

作者： Lee, Wangjin Choi, Jinwook Interdisciplinary Program for Bioengineering Seoul National University Korea Republic of Department of Biomedical Engineering College of Medicine Seoul National University Korea Republic of Institute of Medical and Biological Engineering Medical Research Center Seoul National University Korea Republic of

This paper presents a method of designing specific high-order dependency factor on the linear chain conditional random fields (CRFs) for named entity recognition (NER). Named entities tend to be separated from each other by multiple outside tokens in a text, and thus the first-order CRF as well as the second-order CRF may innately lose transition information between distant named entities. The proposed design uses outside label in NER as a transmission medium of precedent entity information on the CRF. Then, empirical results apparently demonstrate that it is possible to exploit long-distance label dependency in the original first-order linear chain CRF structure upon NER while reducing computational loss rather than in the second-order CRF. Copyright © 2018, The Authors. All rights reserved.

关键词： Random processes

来源：评论

学校读者我要写书评

暂无评论

Effect of sensor configurations on indirect-contact photoplethysmogram measurement system

Effect of sensor configurations on indirect-contact photople...

引用

Cairo International Biomedical Engineering Conference, CIBEC

作者： Hyun Jae Baek Soo Young Sim Jung Soo Kim Kwang Suk Park Graduate Program in Bioengineering and Advanced Biometric Research Center Seoul National University South Korea Samsung Group Interdisciplinary Program in Medical and Biological Engineering Seoul National University South Korea Department of Biomedical Engineering College of Medicine and Advanced Biometric Research Center Seoul National University South Korea

Indirect-contact photoplethysmography is a new approach to measure photoplethysmogrm over clothing for unconstrained and unconscious monitoring purpose. In this paper, the effect of sensor configurations on indirect-contact photoplehysmography was investigated. Sensor configuration was classified into two groups. One was multiple light sources with centered photo-detector configuration and the other was multiple photo-detectors with centered light source configuration. Total 8 kinds of sensors were tested. Experimental results showed that although both sensor groups have possibility to photoplethysmogram measurement over clothing, sensors with multiple light sources showed better performance when thick clothes were considered. Sensor configuration should be selected with regard to its application considerations such as power consumption or target subjects.

关键词： Light sources Light emitting diodes Photonics Biomedical measurements Cotton Clothing Thickness measurement

来源：评论

学校读者我要写书评

暂无评论

Generation of multi-tracer PET attenuation maps using zero-shot learning

Generation of multi-tracer PET attenuation maps using zero-s...

引用

IEEE Symposium on Nuclear Science (NSS/MIC)

作者： M. J. Cho H. S. Shim J. S. Lee Interdisciplinary Program of Bioengineering Seoul National University Seoul Republic of Korea Graduate School Integrated Major in Innovative Medical Science Seoul National University Seoul Republic of Korea Department of Nuclear Medicine Seoul National University Seoul Republic of Korea

Attenuation correction is crucial for achieving high accuracy in PET images. Unfortunately, attenuation correction cannot be performed on PET images when transmission or CT data is unavailable. In such cases, an attenuation map is obtained by utilizing additional anatomical or morphological images. In previous studies, specific models for each tracer in PET images were trained to generate attenuation maps. However, this approach is associated with two major limitations. First, computational resources are greatly increased as each tracer requires a separate network for training. Second, PET images with less commonly used tracers usually have a shortage of data, making it challenging to train a network. To address these limitations, we propose a zero-shot learning method for generating CT-derived attenuation maps (µ-CT) from MLAA-derived attenuation maps (µ-MLAA) using only 18 F-FDG images. Unlike tracer-specific methods which require a separate network for each tracer, this method utilizes a single generator network that encodes the structural features of the input image and the style of the reference µCT to generate the attenuation map. We evaluated our method using clinical brain PET/CT scan datasets of three different tracers, which are 18 F-FDG, 18 F-Florbetaben, and 18 F-FP-CIT. The proposed method which was exclusively trained on 18 F-FDG performs better than tracer-specific networks trained on the same tracer across all evaluation metrics. Moreover, the proposed method generates predictions for the tracers 18 F-Florbetaben and 18 F-FP-CIT, which were not included in the training data, that are either highly comparable or not significantly different from those produced by the tracer-specific networks trained on the corresponding tracer data. The results demonstrate the effectiveness of our zero-shot learning approach in generating multi-tracer PET attenuation maps with only 18 F-FDG.

关键词：

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：