检索结果-内蒙古大学图书馆

Short: Prediction of fetal blood oxygen content in response to partial occlusion of maternal aorta

Smart Health 2023年 28卷 100391页

作者： Qian, Weitai Zhong, Hongtao Ghiasi, Soheil Department of Electrical and Computer Engineering University of California Davis Davis 95618 CA United States

Acute hemorrhage in pregnancy may lead to maternal and/or fetal morbidity or mortality. In emergency medicine, blockage of the aorta via an inflatable endovascular balloon, technically referred to Resuscitative Endovascular Balloon Occlusion of the Aorta (REBOA), is used to manage hemorrhage. However, the application of REBOA in pregnancy needs to strike a balance between two competing objectives of limiting maternal blood loss and ensuring fetal wellness, for which one would need to predict the impact of regulated blood pressure on fetal wellness. To address this problem, we propose an efficient machine learning-based method to predict the temporal impact of the distal Mean Arterial Blood Pressure (dMAP) controlled by the REBOA on the oxygen content in the fetal blood. Evaluation of the algorithm on data collected from in-vivo experiments from pregnant ewe animal models exhibits mean absolute error of 0.61, 1.09, 1.42, 1.70 mmHg, and coefficient of determination of 0.95, 0.86, 0.76, 0.64 for prediction of partial pressure of oxygen in fetal arterial blood, a key predictor of fetal wellness, in 2.5, 5, 7.5, 10-minute prediction horizons, respectively. © 2023 Elsevier Inc.

关键词： autoencoder Fetal assessment Intrapartum hemorrhage Machine learning Resuscitative Endovascular Balloon Occlusion of the Aorta (REBOA) Time-series representation

来源：评论

学校读者我要写书评

暂无评论

Inferring single-cell transcriptomic dynamics with structured latent gene expression dynamics

引用

CELL REPORTS METHODS 2023年第9期3卷 100581页

作者： Farrell, Spencer Mani, Madhav Goyal, Sidhartha Univ Toronto Dept Phys Toronto ON M5S1A7 Canada Northwestern Univ Dept Engn Sci & Appl Math Evanston IL 60208 USA Northwestern Univ Simons Ctr Quantitat Biol NSF Evanston IL 60208 USA Northwestern Univ Dept Mol Biosci Evanston IL 60208 USA Univ Toronto Inst Biomed Engn Toronto ON M5S3G9 Canada

Gene expression dynamics provide directional information for trajectory inference from single-cell RNA sequencing data. Traditional approaches compute RNA velocity using strict modeling assumptions about transcription and splicing of RNA. This can fail in scenarios where multiple lineages have distinct gene dy-namics or where rates of transcription and splicing are time dependent. We present "LatentVelo,"an approach to compute a low-dimensional representation of gene dynamics with deep learning. LatentVelo embeds cells into a latent space with a variational autoencoder and models differentiation dynamics on this "dynamics-based"latent space with neural ordinary differential equations. LatentVelo infers a latent reg-ulatory state that controls the dynamics of an individual cell to model multiple lineages. LatentVelo can pre-dict latent trajectories, describing the inferred developmental path for individual cells rather than just local RNA velocity vectors. The dynamics-based embedding batch corrects cell states and velocities, outperform-ing comparable autoencoder batch correction methods that do not consider gene expression dynamics.

关键词： RNA velocity trajectory inference batch correction autoencoder cell-fate transitions deep learning representation learning neural ODE CP: Systems biology

来源：评论

学校读者我要写书评

暂无评论

AE-ACG: A novel deep learning-based method for stock price movement prediction

引用

FINANCE RESEARCH LETTERS 2023年第PartA期58卷

作者： Li, Shicheng Huang, Xiaoyong Cheng, Zhonghou Zou, Wei Yi, Yugen Jiangxi Normal Univ Res Ctr Management Sci & Engn Nanchang 330022 Peoples R China Jiangxi Normal Univ Sch Finance Nanchang 330022 Peoples R China Jiangxi Normal Univ Sch Software Nanchang 330022 Peoples R China

This paper proposes a method named AE-ACG for stock price movement prediction. In AE-ACG, the convolutional neural network (CNN) and gated recurrent unit (GRU) are combined to design a base layer, which is embedded in the autoencoder (AE) framework, to efficiently extract features from financial time series data. Furthermore, skip connection links encoding and decoding to leverage hierarchical features. Attention mechanism (AM) also distinguishes the importance of historical data across periods. Extensive experiments demonstrated that the proposed model is effective in predicting price movements, showing advantages over some mainstream methods.

关键词： Deep learning Financial time series forecasting Stock price movement autoencoder

来源：评论

学校读者我要写书评

暂无评论

Early Glaucoma Detection by Using Style Transfer to Predict Retinal Nerve Fiber Layer Thickness Distribution on the Fundus Photograph

引用

OPHTHALMOLOGY SCIENCE 2022年第3期2卷 100180-100180页

作者： Chen, Henry Shen-Lih Chen, Guan-An Syu, Jhen-Yang Chuang, Lan-Hsin Su, Wei -Wen Wu, Wei -Chi Liu, Jian-Hong Chen, Jian-Ren Huang, Su -Chen Kang, Eugene Yu-Chuan Chang Gung Mem Hosp Linkou Med Ctr Dept Ophthalmol Taoyuan Taiwan Chang Gung Univ Coll Med Taoyuan Taiwan Ind Technol Res Inst Serv Syst Technol Ctr Dept Intelligent Med & Healthcare Healthcare Serv Div Hsinchu Taiwan Keelung Chang Gung Mem Hosp Dept Ophthalmol Keelung Taiwan Chang Gung Univ Grad Inst Clin Med Sci Coll Med Taoyuan Taiwan Chang Gung Mem Hosp Dept Ophthalmol 5 Fu-Hsin Rd Taoyuan 333 Taiwan

Objective: We aimed to develop a deep learning (DL) -based algorithm for early glaucoma detection based on color fundus photographs that provides information on defects in the retinal nerve fiber layer (RNFL) and its thickness from the mapping and translating relations of spectral domain OCT (SD-OCT) thickness ***: Developing and evaluating an artificial intelligence detection ***: Pretraining paired data of color fundus photographs and SD-OCT images from 189 healthy par-ticipants and 371 patients with early glaucoma were ***: The variational autoencoder (VAE) network training architecture was used for training, and the correlation between the fundus photographs and RNFL thickness distribution was determined through the deep neural network. The reference standard was defined as a vertical cup-to-disc ratio of >0.7, other typical changes in glaucomatous optic neuropathy, and RNFL defects. Convergence indicates that the VAE has learned a dis-tribution that would enable us to produce corresponding synthetic OCT *** Outcome Measures: Similarly to wide-field OCT scanning, the proposed model can extract the results of RNFL thickness analysis. The structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) were used to assess signal strength and the similarity in the structure of the color fundus images converted to an RNFL thickness distribution model. The differences between the model-generated images and original images were quantified. Results: We developed and validated a novel DL-based algorithm to extract thickness information from the color space of fundus images similarly to that from OCT images and to use this information to regenerate RNFL thickness distribution images. The generated thickness map was sufficient for clinical glaucoma detection, and the generated images were similar to ground truth (PSNR: 19.31 decibels;SSIM: 0.44). The inference results were similar to the OCT-generated original images in te

关键词： Image-to-image translation Glaucoma autoencoder OCT Retinal nerve fiber layer thickness

来源：评论

学校读者我要写书评

暂无评论

iSubGen generates integrative disease subtypes by pairwise similarity assessment

引用

CELL REPORTS METHODS 2024年第11期4卷 100884页

作者： Fox, Natalie S. Tian, Mao Markowitz, Alexander L. Haider, Syed Li, Constance H. Boutros, Paul C. Univ Toronto Dept Med Biophys Toronto ON M5G 1L7 Canada Univ Calif Los Angeles Dept Human Genet Los Angeles CA 90095 USA Univ Calif Los Angeles Inst Precis Hlth Los Angeles CA 90095 USA Univ Calif Los Angeles Jonsson Comprehens Canc Ctr Los Angeles CA 90095 USA Ontario Inst Canc Res Toronto ON M5G 0A3 Canada Inst Canc Res Breast Canc Now Toby Robins Res Ctr London England Univ Toronto Dept Pharmacol & Toxicol Toronto ON M5S 1A8 Canada Univ Calif Los Angeles Dept Urol Los Angeles CA 90095 USA Univ Calif Los Angeles Broad Stem Cell Res Ctr Los Angeles CA 90095 USA

There are myriad types of biomedical data-molecular, clinical images, and others. When a group of patients with the same underlying disease exhibits similarities across multiple types of data, this is called a subtype. Existing subtyping approaches struggle to handle diverse data types with missing information. To improve subtype discovery, we exploited changes in the correlation-structure between different data types to create iSubGen, an algorithm for integrative subtype generation. iSubGen can accommodate any feature that can be compared with a similarity metric to create subtypes versatilely. It can combine arbitrary data types for subtype discovery, such as merging genetic, transcriptomic, proteomic, and pathway data. iSubGen recapitulates known subtypes across multiple cancers even with substantial missing data and identifies subtypes with distinct clinical behaviors. It performs equally with or superior to other subtyping methods, offering greater stability and robustness to missing data and flexibility to new data types. It is available at https:// ***/web/packages/iSubGen.

关键词： subtype discovery cancer biology biomarkers algorithm pattern discovery multi-omics data integration autoencoder system biology data correlation

来源：评论

学校读者我要写书评

暂无评论

Encoding and exploring latent design space of optimal material structures via a VAE-LSTM model

引用

FORCES IN MECHANICS 2021年 5卷

作者： Lew, Andrew J. Buehler, Markus J. MIT Lab Atomist & Mol Mech LAMM 77 Massachusetts Ave Cambridge MA 02139 USA MIT Dept Chem 77 Massachusetts Ave Cambridge MA 02139 USA MIT Schwarzman Coll Comp Ctr Computat Sci & Engn 77 Massachusetts Ave Cambridge MA 02139 USA

Variational autoencoders (VAE) are machine learning models that can extract low dimensional representations of data from datasets of high complexity and volume. Importantly, they can be used for generative purposes to reconstruct complex data, such as images, from a low dimensional encoding of only a few variables. Long shortterm memory (LSTM) neural networks are well suited to learning logical trajectory relationships within datasets. Using these two models in concert, we develop a VAE-LSTM approach to learn a classic mechanical materials design problem. Here, we focus on the compliance optimization of cantilever design, using a VAE to encode cantilever structures into a 2D latent space and a LSTM to learn trajectories in that latent space corresponding to the optimization process. Ultimately, we are able to clearly visualize the space of cantilever design, generate new design with extremely low density beyond the original dataset, and obtain optimal cantilever structures inspired by nature. We also demonstrate how the resulting designs can be manufactured using 3D printing, offering a rapid pathway from concept to prototype. The method we developed here can be generalized to other imagebased datasets encapsulating changes from multiple factors. The ability offered by our approach to interpret complex behavior, via representations in simplified space, has great potential for application in the intelligent design and manufacturing of materials structure problems.

关键词： Mechanics Design Machine learning 3D printing autoencoder Topology optimization Structural design Optimization Deep learning AI

来源：评论

学校读者我要写书评

暂无评论

NUC-Fuse: Multimodal medical image fusion using nuclear norm & classification of brain tumors using ARBFN

引用

Intelligence-Based Medicine 2024年 10卷 100181-100181页

作者： H., Shihabudeen J., Rajeesh College of Engineering Thalassery APJ Abdul Kalam Technological University India Department of Electronics College of Engineering Perumon Kerala India

Medical imaging has been widely used to diagnose diseases over the past two decades. The lack of information in this field makes it difficult for medical experts to diagnose diseases with a single modality. The combination of image fusion techniques enables the integration of pictures depicting various tissues and disorders from multiple medical imaging devices, facilitating enhanced research and treatment by providing complementary information through multimodal medical imaging fusion. The proposed work employs the nuclear norm and residual connections to combine the complementary features from both CT and MRI imaging approaches. The autoencoder eventually creates a merged image. The fused pictures are categorized as benign or malignant in the following phase using the present Radial Basis Function Network (RBFN). The performance measures, such as Mutual Information, Structural Similarity Index Measure, Q_w, and Q_e, have shown improved values, specifically 4.6328, 0.6492, 0.8300, and 0.8185 respectively, when compared with different fusion methods. Additionally, the classification algorithm yields 97% accuracy, 89% precision, and 92% recall when combined with the proposed fusion algorithm. © 2024 The Authors

关键词： autoencoder Classification Image fusion Imaging modality Nuclear norm

来源：评论

学校读者我要写书评

暂无评论

Comparison of tabular synthetic data generation techniques using propensity and cluster log metric

引用

International Journal of Information Management Data Insights 2023年第2期3卷

作者： Pathare, Aryan Mangrulkar, Ramchandra Suvarna, Kartik Parekh, Aryan Thakur, Govind Gawade, Aruna Computer Engineering Dwarkadas J. Sanghvi College of Engineering Maharashtra Mumbai 400056 India

In the 21st-century, data is as valuable as gold. Many data-centric applications are generating a vast amount of data. Businesses can use this generated data to pinpoint the various sources of problems, if any. In addition, the data can help enterprises to identify connections between what is happening in different areas, departments, and systems. However, having more data is not enough;the data should also be of high quality. For example, taking action based on unfamiliar evidence, speculative ideas, or observations could lead to the wastage of resources. Whereas using high-quality data will help achieve correct results. Synthetic data is artificially generated data. Synthetic data is generated by an algorithm and used to represent real-world data, test datasets, perform mathematical model validation, and, most importantly, for training of machine learning models. Synthetic data can also be used to preserve data privacy. It is considered a safe way to transfer sensitive data because it creates a transaction database that does not contain any confidential information. This paper compares the tabular synthetic data generation techniques using various datasets, viz. balanced datasets, unbalanced datasets, datasets with numerical attributes only, datasets with categorical attributes only and mixed datasets. The utility of the generated synthetic data is measured using the Propensity score metric and Cluster-Log metric. The main finding of this paper is that the Classification And Regression Tree (CART) model provides the best results for all types of datasets. At the same time, Generative Adversarial Networks (GANs) give subpar or mediocre results at best. This contradicts the common belief that GANs are the go-to models for producing synthetic data. © 2023 The Author(s)

关键词： autoencoder Bayesian network CART Copula GAN RNN Synthetic data TVAE

来源：评论

学校读者我要写书评

暂无评论

基于集成深度学习算法的燃气负荷预测方法

引用

计算机系统应用 2019年第12期28卷 47-54页

作者：王晓霞徐晓钟张彤高超伟上海师范大学信息与机电工程学院

燃气负荷预测受到社会经济、天气因素、日期类型等多种复杂因素的影响,而多因素的共同作用则必然会导致燃气负荷序列变化趋势具有很大的随机性以及一定程度上的复杂性.为了有效提高燃气负荷预测的精度,本文提出了一种新型的集成深度算... 详细信息

燃气负荷预测受到社会经济、天气因素、日期类型等多种复杂因素的影响,而多因素的共同作用则必然会导致燃气负荷序列变化趋势具有很大的随机性以及一定程度上的复杂性.为了有效提高燃气负荷预测的精度,本文提出了一种新型的集成深度算法来对燃气负荷进行多步预测.首先通过EEMD算法将非平稳非线性的负荷序列分解为若干个稳态且线性的本征模式分量及剩余项,有效的避免了传统EMD带来的模态混叠问题,然后将负荷数据的影响因素输入到autoencoder中进行特征提取并做非线性降维处理,再将EEMD分解得到的每个子序列分别与autoencoder提取到的特征序列组成不同的训练矩阵,最后针对不同的子序列对应的训练矩阵建立相应的LSTM预测模型,重构分量预测值得到最终预测结果.为了验证所提出算法的有效性和预测性能,使用上海燃气数据来进行上述模型的仿真实验,结果证明相较对比方法,预测精度有了明显的提高.

关键词：燃气负荷预测长短时记忆网络集合经验模式分解集成算法 autoencoder 时间序列分析

来源：评论

学校读者我要写书评

暂无评论

基於協同神經網路與彙整機制建構零時差攻擊偵測系統

引用

资讯安全通讯 2020年第1期26卷 8-24页

作者：陳志龍魏國瑞陳映親李榮三

基於特徵检测的入侵侦测系统在防护网路攻击中扮演着十分重要的角色，然而针对已知特徵进行检验的做法，却存在无法侦测零时差攻击之缺陷，造成入侵侦测系统容易忽略新型恶意行为，进而无法有效地分辨正常与恶意流量，导致企业或个人资... 详细信息

基於特徵检测的入侵侦测系统在防护网路攻击中扮演着十分重要的角色，然而针对已知特徵进行检验的做法，却存在无法侦测零时差攻击之缺陷，造成入侵侦测系统容易忽略新型恶意行为，进而无法有效地分辨正常与恶意流量，导致企业或个人资料外泄而造成极大的损失。在本篇论文中，我们结合autoencoder以及深层神经网路，提出可检测未知攻击的入侵侦测系统，不仅可检验已知的恶意行为，亦可改善无法抵御零时差攻击之缺点。在本系统架构中，零时差攻击侦测模组负责辨别收集的流量是否有未知攻击的出现；而已知攻击分类方法则作为攻击类型的分类器以进一步判断流量具体属於何种已知攻击。接着，为使系统学习未知攻击并将其转换为已知攻击，我们加入攻击汇整机制，透过基於DBSCAN的具投票制度之分群法，将特徵相近的未知攻击聚合成新型态的攻击类别，使系统的攻击侦测能力可随着恶意攻击的数量而不断成长。实验结果表明，本论文所提出的新型态入侵侦测系统能有效的侦测未知攻击，并具备优良的分类结果。

关键词：入侵偵測系統零時差攻擊 autoencoder 彙整機制

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：