Federated learning(FL),as a distributed learning paradigm,allows multiple medical institutions to collaborate on learning without the need to centralize all client data.However,existing methods pay little attention to...Federated learning(FL),as a distributed learning paradigm,allows multiple medical institutions to collaborate on learning without the need to centralize all client data.However,existing methods pay little attention to more challenging medical image semantic segmentation tasks,especially in the scenario of the imbalanced dataset in federated few-shot learning(FSL).In this paper,we propose a subnetwork-based federated few-shot organ image segmentation method.Firstly,individual clients train using local training samples and then upload local model gradients to the server.The server utilizes their respective local model gradients to update the subnetwork maintained on the server and generate aggregation weights for forming personalized model parameters.Through this method,we can learn the similarities between different clients to address data heterogeneity issues.In addition,to enhance the communication efficiency between clients and the server,we have also designed a personalized layer aggregation strategy,which only transmits partial layer model parameters during the communication process to improve communication efficiency.Finally,we conducted experiments on abdomen magnetic resonance imaging(ABD-MRI)and abdomen computed tomography(ABD-CT)datasets to demonstrate the effectiveness of our method.展开更多
Patients with mild traumatic brain injury have a diverse clinical presentation,and the underlying pathophysiology remains poorly understood.Magnetic resonance imaging is a non-invasive technique that has been widely u...Patients with mild traumatic brain injury have a diverse clinical presentation,and the underlying pathophysiology remains poorly understood.Magnetic resonance imaging is a non-invasive technique that has been widely utilized to investigate neuro biological markers after mild traumatic brain injury.This approach has emerged as a promising tool for investigating the pathogenesis of mild traumatic brain injury.G raph theory is a quantitative method of analyzing complex networks that has been widely used to study changes in brain structure and function.However,most previous mild traumatic brain injury studies using graph theory have focused on specific populations,with limited exploration of simultaneous abnormalities in structural and functional connectivity.Given that mild traumatic brain injury is the most common type of traumatic brain injury encounte red in clinical practice,further investigation of the patient characteristics and evolution of structural and functional connectivity is critical.In the present study,we explored whether abnormal structural and functional connectivity in the acute phase could serve as indicators of longitudinal changes in imaging data and cognitive function in patients with mild traumatic brain injury.In this longitudinal study,we enrolled 46 patients with mild traumatic brain injury who were assessed within 2 wee ks of injury,as well as 36 healthy controls.Resting-state functional magnetic resonance imaging and diffusion-weighted imaging data were acquired for graph theoretical network analysis.In the acute phase,patients with mild traumatic brain injury demonstrated reduced structural connectivity in the dorsal attention network.More than 3 months of followup data revealed signs of recovery in structural and functional connectivity,as well as cognitive function,in 22 out of the 46 patients.Furthermore,better cognitive function was associated with more efficient networks.Finally,our data indicated that small-worldness in the acute stage could serve as a predictor of longitudinal changes in connectivity in patients with mild traumatic brain injury.These findings highlight the importance of integrating structural and functional connectivity in unde rstanding the occurrence and evolution of mild traumatic brain injury.Additionally,exploratory analysis based on subnetworks could serve a predictive function in the prognosis of patients with mild traumatic brain injury.展开更多
The problem for the solvability of pseudo-tearing subnetwork is one of the essentialinvestigations of network theory.The results presented would be not only mathematical conditionsbut also topological conditions for s...The problem for the solvability of pseudo-tearing subnetwork is one of the essentialinvestigations of network theory.The results presented would be not only mathematical conditionsbut also topological conditions for subnetwork solvability.These conditions are necessary andalmost sufficient.It should guide one intuitively to the design of accessible nodes.展开更多
引文网络中的学术群体识别旨在揭示学者间的内在关联。然而,节点属性(内容特征)与拓扑结构(关联模式)的异质性导致融合时易出现特征错位与信息稀释,且现有方法常忽略由高连接度节点构成的核心子网络所蕴含的稳定关联模式,这些制约了识...引文网络中的学术群体识别旨在揭示学者间的内在关联。然而,节点属性(内容特征)与拓扑结构(关联模式)的异质性导致融合时易出现特征错位与信息稀释,且现有方法常忽略由高连接度节点构成的核心子网络所蕴含的稳定关联模式,这些制约了识别精度。为了解决上述问题,提出一种多视图注意力融合的学术群体识别(MAFCI)方法。所提方法采用双视图设计:一个视图对原始节点属性与邻接矩阵进行编码,另一个视图融合由k-core提取的核心节点信息及其拓扑结构;采用稀疏图注意力机制分别从两个视图中提取低维嵌入,并通过动态加权策略实现多源信息的自适应融合,以兼顾局部结构与全局语义。解码器通过内积重构邻接矩阵,并基于注意力机制恢复节点属性与核心信息;同时,引入自监督聚类损失,引导嵌入表示向判别性聚类中心收敛,增强群体边界的区分度。在Cora、Citeseer、ACM和DBLP 4个引文网络数据集上的实验结果表明,MAFCI方法在多数核心评价指标上优于DDGAE(Deep Dual Graph Attention auto-Encoder)和BCDAN(Balanced method for Community Detection in Attribute Networks)等9个先进基线方法。具体地,在DBLP数据集上,MAFCI方法的归一化互信息(NMI)达到0.513,准确率(ACC)达到0.809,分别较最优基线方法提升了2.3和1.3个百分点;在ACM数据集上,MAFCI方法的NMI为0.674,F1分数为0.903,分别较最优基线方法提升了1.2和0.8个百分点;在Citeseer数据集上,MAFCI方法的NMI达到0.448,较最优基线方法提升了0.7个百分点;在Cora数据集上,MAFCI方法的NMI达到0.543,同样优于所有对比方法。可见,MAFCI方法通过双视图稀疏注意力与动态融合机制,有效提升了学术群体识别的准确性与鲁棒性,在不同规模与结构的引文网络上均表现出良好泛化能力。展开更多
Time delay and coupling strength are important factors that affect the synchronization of neural networks.In this study,a modular neural network containing subnetworks of different scales was constructed using the Hod...Time delay and coupling strength are important factors that affect the synchronization of neural networks.In this study,a modular neural network containing subnetworks of different scales was constructed using the Hodgkin–Huxley(HH)neural model;i.e.,a small-scale random network was unidirectionally connected to a large-scale small-world network through chemical synapses.Time delays were found to induce multiple synchronization transitions in the network.An increase in coupling strength also promoted synchronization of the network when the time delay was an integer multiple of the firing period of a single neuron.Considering that time delays at different locations in a modular network may have different effects,we explored the influence of time delays within each subnetwork and between two subnetworks on the synchronization of modular networks.We found that when the subnetworks were well synchronized internally,an increase in the time delay within both subnetworks induced multiple synchronization transitions of their own.In addition,the synchronization state of the small-scale network affected the synchronization of the large-scale network.It was surprising to find that an increase in the time delay between the two subnetworks caused the synchronization factor of the modular network to vary periodically,but it had essentially no effect on the synchronization within the receiving subnetwork.By analyzing the phase difference between the two subnetworks,we found that the mechanism of the periodic variation of the synchronization factor of the modular network was the periodic variation of the phase difference.Finally,the generality of the results was demonstrated by investigating modular networks at different scales.展开更多
The active-subnetwork-extraction theorem and the passive-subnetwork-extraction theorem are derived from one of the author’s previous work. Using them to find the fully symbolic network functions, the multilevel-teari...The active-subnetwork-extraction theorem and the passive-subnetwork-extraction theorem are derived from one of the author’s previous work. Using them to find the fully symbolic network functions, the multilevel-tearing topological analysis for active networks can be substantially simplified so that it can be conducted conveniently on a computer. Using them to find partially symbolic network functions, one finds it possible to extend not only the electrical network that can be analysed on a computer to the order that can be processed by an ordinary numerical analysis program, but also the symbolic subnetwork to the order that can be processed by an ordinary topological analysis program. So far the latter cannot be achieved with the current conventional methods, i. e. the parameter-extraction method and interpolative approach.展开更多
Understanding the genetic mechanisms underlying particular adaptations/phenotypes of organisms is one of the core issues of evolutionary biology.The use of genomic data has greatly advanced our understandings on this ...Understanding the genetic mechanisms underlying particular adaptations/phenotypes of organisms is one of the core issues of evolutionary biology.The use of genomic data has greatly advanced our understandings on this issue,as well as other aspects of evolutionary biology,including molecular adaptation,speciation,and even conservation of endangered species.Despite the well-recognized advantages,usages of genomic data are still limited to non-mammal vertebrate groups,partly due to the difficulties in assembling large or highly heterozygous genomes.Although this is particularly the case for amphibians,nonetheless,several comparative and population genomic analyses have shed lights into the speciation and adaptation processes of amphibians in a complex landscape,giving a promising hope for a wider application of genomics in the previously believed challenging groups of organisms.At the same time,these pioneer studies also allow us to realize numerous challenges in studying the molecular adaptations and/or phenotypic evolutionary mechanisms of amphibians.In this review,we first summarize the recent progresses in the study of adaptive evolution of amphibians based on genomic data,and then we give perspectives regarding how to effectively identify key pathways underlying the evolution of complex traits in the genomic era,as well as directions for future research.展开更多
Human motion recognition plays a crucial role in the video analysis framework.However,a given video may contain a variety of noises,such as an unstable background and redundant actions,that are completely different fr...Human motion recognition plays a crucial role in the video analysis framework.However,a given video may contain a variety of noises,such as an unstable background and redundant actions,that are completely different from the key actions.These noises pose a great challenge to human motion recognition.To solve this problem,we propose a new method based on the 3-Dimensional(3D)Bag of Visual Words(BoVW)framework.Our method includes two parts:The first part is the video action feature extractor,which can identify key actions by analyzing action features.In the video action encoder,by analyzing the action characteristics of a given video,we use the deep 3D CNN pre-trained model to obtain expressive coding information.A classifier with subnetwork nodes is used for the final classification.The extensive experiments demonstrate that our method leads to an impressive effect on complex video analysis.Our approach achieves state-of-the-art performance on the datasets of UCF101(85.3%)and HMDB51(54.5%).展开更多
基金supported by the National Natural Science Foundation of China(No.61713447)。
摘要Federated learning(FL),as a distributed learning paradigm,allows multiple medical institutions to collaborate on learning without the need to centralize all client data.However,existing methods pay little attention to more challenging medical image semantic segmentation tasks,especially in the scenario of the imbalanced dataset in federated few-shot learning(FSL).In this paper,we propose a subnetwork-based federated few-shot organ image segmentation method.Firstly,individual clients train using local training samples and then upload local model gradients to the server.The server utilizes their respective local model gradients to update the subnetwork maintained on the server and generate aggregation weights for forming personalized model parameters.Through this method,we can learn the similarities between different clients to address data heterogeneity issues.In addition,to enhance the communication efficiency between clients and the server,we have also designed a personalized layer aggregation strategy,which only transmits partial layer model parameters during the communication process to improve communication efficiency.Finally,we conducted experiments on abdomen magnetic resonance imaging(ABD-MRI)and abdomen computed tomography(ABD-CT)datasets to demonstrate the effectiveness of our method.
基金supported by the National Natural Science Foundation of China,Nos.81671671(to JL),61971451(to JL),U22A2034(to XK),62177047(to XK)the National Defense Science and Technology Collaborative Innovation Major Project of Central South University,No.2021gfcx05(to JL)+6 种基金Clinical Research Cen terfor Medical Imaging of Hunan Province,No.2020SK4001(to JL)Key Emergency Project of Pneumonia Epidemic of Novel Coronavirus Infection of Hu nan Province,No.2020SK3006(to JL)Innovative Special Construction Foundation of Hunan Province,No.2019SK2131(to JL)the Science and Technology lnnovation Program of Hunan Province,Nos.2021RC4016(to JL),2021SK53503(to ML)Scientific Research Program of Hunan Commission of Health,No.202209044797(to JL)Central South University Research Program of Advanced Interdisciplinary Studies,No.2023Q YJC020(to XK)the Natural Science Foundation of Hunan Province,No.2022JJ30814(to ML)。
摘要Patients with mild traumatic brain injury have a diverse clinical presentation,and the underlying pathophysiology remains poorly understood.Magnetic resonance imaging is a non-invasive technique that has been widely utilized to investigate neuro biological markers after mild traumatic brain injury.This approach has emerged as a promising tool for investigating the pathogenesis of mild traumatic brain injury.G raph theory is a quantitative method of analyzing complex networks that has been widely used to study changes in brain structure and function.However,most previous mild traumatic brain injury studies using graph theory have focused on specific populations,with limited exploration of simultaneous abnormalities in structural and functional connectivity.Given that mild traumatic brain injury is the most common type of traumatic brain injury encounte red in clinical practice,further investigation of the patient characteristics and evolution of structural and functional connectivity is critical.In the present study,we explored whether abnormal structural and functional connectivity in the acute phase could serve as indicators of longitudinal changes in imaging data and cognitive function in patients with mild traumatic brain injury.In this longitudinal study,we enrolled 46 patients with mild traumatic brain injury who were assessed within 2 wee ks of injury,as well as 36 healthy controls.Resting-state functional magnetic resonance imaging and diffusion-weighted imaging data were acquired for graph theoretical network analysis.In the acute phase,patients with mild traumatic brain injury demonstrated reduced structural connectivity in the dorsal attention network.More than 3 months of followup data revealed signs of recovery in structural and functional connectivity,as well as cognitive function,in 22 out of the 46 patients.Furthermore,better cognitive function was associated with more efficient networks.Finally,our data indicated that small-worldness in the acute stage could serve as a predictor of longitudinal changes in connectivity in patients with mild traumatic brain injury.These findings highlight the importance of integrating structural and functional connectivity in unde rstanding the occurrence and evolution of mild traumatic brain injury.Additionally,exploratory analysis based on subnetworks could serve a predictive function in the prognosis of patients with mild traumatic brain injury.
基金This project supported by National Natural Science Foundation of China
摘要The problem for the solvability of pseudo-tearing subnetwork is one of the essentialinvestigations of network theory.The results presented would be not only mathematical conditionsbut also topological conditions for subnetwork solvability.These conditions are necessary andalmost sufficient.It should guide one intuitively to the design of accessible nodes.
摘要引文网络中的学术群体识别旨在揭示学者间的内在关联。然而,节点属性(内容特征)与拓扑结构(关联模式)的异质性导致融合时易出现特征错位与信息稀释,且现有方法常忽略由高连接度节点构成的核心子网络所蕴含的稳定关联模式,这些制约了识别精度。为了解决上述问题,提出一种多视图注意力融合的学术群体识别(MAFCI)方法。所提方法采用双视图设计:一个视图对原始节点属性与邻接矩阵进行编码,另一个视图融合由k-core提取的核心节点信息及其拓扑结构;采用稀疏图注意力机制分别从两个视图中提取低维嵌入,并通过动态加权策略实现多源信息的自适应融合,以兼顾局部结构与全局语义。解码器通过内积重构邻接矩阵,并基于注意力机制恢复节点属性与核心信息;同时,引入自监督聚类损失,引导嵌入表示向判别性聚类中心收敛,增强群体边界的区分度。在Cora、Citeseer、ACM和DBLP 4个引文网络数据集上的实验结果表明,MAFCI方法在多数核心评价指标上优于DDGAE(Deep Dual Graph Attention auto-Encoder)和BCDAN(Balanced method for Community Detection in Attribute Networks)等9个先进基线方法。具体地,在DBLP数据集上,MAFCI方法的归一化互信息(NMI)达到0.513,准确率(ACC)达到0.809,分别较最优基线方法提升了2.3和1.3个百分点;在ACM数据集上,MAFCI方法的NMI为0.674,F1分数为0.903,分别较最优基线方法提升了1.2和0.8个百分点;在Citeseer数据集上,MAFCI方法的NMI达到0.448,较最优基线方法提升了0.7个百分点;在Cora数据集上,MAFCI方法的NMI达到0.543,同样优于所有对比方法。可见,MAFCI方法通过双视图稀疏注意力与动态融合机制,有效提升了学术群体识别的准确性与鲁棒性,在不同规模与结构的引文网络上均表现出良好泛化能力。
基金supported by the National Natural Science Foundation of China(No.12175080)the Fundamental Research Funds for the Central Universities,China(No.CCNU22JC009)。
摘要Time delay and coupling strength are important factors that affect the synchronization of neural networks.In this study,a modular neural network containing subnetworks of different scales was constructed using the Hodgkin–Huxley(HH)neural model;i.e.,a small-scale random network was unidirectionally connected to a large-scale small-world network through chemical synapses.Time delays were found to induce multiple synchronization transitions in the network.An increase in coupling strength also promoted synchronization of the network when the time delay was an integer multiple of the firing period of a single neuron.Considering that time delays at different locations in a modular network may have different effects,we explored the influence of time delays within each subnetwork and between two subnetworks on the synchronization of modular networks.We found that when the subnetworks were well synchronized internally,an increase in the time delay within both subnetworks induced multiple synchronization transitions of their own.In addition,the synchronization state of the small-scale network affected the synchronization of the large-scale network.It was surprising to find that an increase in the time delay between the two subnetworks caused the synchronization factor of the modular network to vary periodically,but it had essentially no effect on the synchronization within the receiving subnetwork.By analyzing the phase difference between the two subnetworks,we found that the mechanism of the periodic variation of the synchronization factor of the modular network was the periodic variation of the phase difference.Finally,the generality of the results was demonstrated by investigating modular networks at different scales.
摘要The active-subnetwork-extraction theorem and the passive-subnetwork-extraction theorem are derived from one of the author’s previous work. Using them to find the fully symbolic network functions, the multilevel-tearing topological analysis for active networks can be substantially simplified so that it can be conducted conveniently on a computer. Using them to find partially symbolic network functions, one finds it possible to extend not only the electrical network that can be analysed on a computer to the order that can be processed by an ordinary numerical analysis program, but also the symbolic subnetwork to the order that can be processed by an ordinary topological analysis program. So far the latter cannot be achieved with the current conventional methods, i. e. the parameter-extraction method and interpolative approach.
基金This study was supported by the National Natural Science Foundation of China(31671326,31871275)the Highlevel Talent Introduction Program of Yunnan University to Y.B.S。
摘要Understanding the genetic mechanisms underlying particular adaptations/phenotypes of organisms is one of the core issues of evolutionary biology.The use of genomic data has greatly advanced our understandings on this issue,as well as other aspects of evolutionary biology,including molecular adaptation,speciation,and even conservation of endangered species.Despite the well-recognized advantages,usages of genomic data are still limited to non-mammal vertebrate groups,partly due to the difficulties in assembling large or highly heterozygous genomes.Although this is particularly the case for amphibians,nonetheless,several comparative and population genomic analyses have shed lights into the speciation and adaptation processes of amphibians in a complex landscape,giving a promising hope for a wider application of genomics in the previously believed challenging groups of organisms.At the same time,these pioneer studies also allow us to realize numerous challenges in studying the molecular adaptations and/or phenotypic evolutionary mechanisms of amphibians.In this review,we first summarize the recent progresses in the study of adaptive evolution of amphibians based on genomic data,and then we give perspectives regarding how to effectively identify key pathways underlying the evolution of complex traits in the genomic era,as well as directions for future research.
摘要Human motion recognition plays a crucial role in the video analysis framework.However,a given video may contain a variety of noises,such as an unstable background and redundant actions,that are completely different from the key actions.These noises pose a great challenge to human motion recognition.To solve this problem,we propose a new method based on the 3-Dimensional(3D)Bag of Visual Words(BoVW)framework.Our method includes two parts:The first part is the video action feature extractor,which can identify key actions by analyzing action features.In the video action encoder,by analyzing the action characteristics of a given video,we use the deep 3D CNN pre-trained model to obtain expressive coding information.A classifier with subnetwork nodes is used for the final classification.The extensive experiments demonstrate that our method leads to an impressive effect on complex video analysis.Our approach achieves state-of-the-art performance on the datasets of UCF101(85.3%)and HMDB51(54.5%).