在低标签率场景下,因监督信息极少,现有的半监督方法在获得令人满意的分类结果方面面临挑战。为此提出1种基于自适应正则化补充交叉熵损失的图卷积网络(graph convolutional network based on adaptive regularization complement with ...在低标签率场景下,因监督信息极少,现有的半监督方法在获得令人满意的分类结果方面面临挑战。为此提出1种基于自适应正则化补充交叉熵损失的图卷积网络(graph convolutional network based on adaptive regularization complement with cross entropy loss,ARCGCN),从损失函数的设计角度出发,通过引入补充熵与图拉普拉斯正则化项,构建适用于标签稀疏数据集并能够抑制标签噪声影响的半监督图卷积网络,从而有效提升模型预测精度与泛化能力。在6个数据集3Sources、Citeseer、GRAZ02、Out-Scene、Noisy-MNIST和NGs上进行实验,通过与9种基线方法对比,证明了ARCGCN性能的有效性和优越性。结果表明:特别在1%的极低标签率下,ARCGCN相较于所有基线方法,其准确率提升尤为显著,充分验证了其处理标签稀疏数据的能力。展开更多
A new concept of labeling called the signed product cordial labeling is introduced and investigated for path graph, cycle graphs, star-K1,n, Bistar-Bn,n, and Some general results on signed product cordial labeling are...A new concept of labeling called the signed product cordial labeling is introduced and investigated for path graph, cycle graphs, star-K1,n, Bistar-Bn,n, and Some general results on signed product cordial labeling are studied.展开更多
X-ray computed tomography(CT),ultrasonography(US)and radionuclide scanning are important clinical methods for evaluating morphology of the kidney.These modalities are also applicable for estimating kidney function wit...X-ray computed tomography(CT),ultrasonography(US)and radionuclide scanning are important clinical methods for evaluating morphology of the kidney.These modalities are also applicable for estimating kidney function with time lapse analysis using proper contrastmedia as may be necessary.In the case of US,it can estimate kidney function based on the measurement of blood flow using the Doppler effect.Formerly,magnetic resonance imaging(MRI)was an inappropriate diagnostic imaging technique for abdominal organs because of their respiratory displacements.However,MRI is now actively used for kidney as well as liver or other parenchymal organs,in tandem with the technological advances.Unlike unenhanced X-ray CT,"conventional"MRI can distinguish the border between cortex and medulla in T1 or T2 weighted images.It was known that the border blurred with decreasing kidney function.Moreover,several other particular imaging methods were introduced in recent years,and these could be called"functional"MRI.In this review,the following are discussed:functional MRI for chronic kidney disease,which include blood oxygenation level-dependent MRI for evaluation of hypoxia,diffusion-weighted imagingfor evaluation of fibrosis,diffusion tensor imaging for evaluation of microstructure,and arterial spin labeling to evaluate the amount of organ perfusion,accompanied with several related articles.The ultimate goal of functional MRI is to provide useful in vivo information repeatedly for daily medical treatment non-invasively.展开更多
Arterial spin labeling(ASL) is a magnetic resonance imaging technique for measuring tissue perfusion using a freely diffusible intrinsic tracer.As compared with other perfusion techniques,ASL offers several advantages...Arterial spin labeling(ASL) is a magnetic resonance imaging technique for measuring tissue perfusion using a freely diffusible intrinsic tracer.As compared with other perfusion techniques,ASL offers several advantages and is now available for routine clinical practice in many institutions.Its noninvasive nature and ability to quantitatively measure tissue perfusion make ASL ideal for research and clinical studies.Recent technical advances have increased its sensitivity and also extended its potential applications.This review focuses on some basic knowledge of ASL perfusion,emerging techniques and clinical applications in neuroimaging.展开更多
目的:评价动脉自旋标记(ASL)对急性缺血性脑卒中(AIS)患者早期神经功能衰退(END)的预测价值。方法:收集198例AIS患者,根据美国国立卫生研究院脑卒中量表(NIHSS)评分分为END组68例和非END组130例。通过单因素和多因素logistic回归分析评...目的:评价动脉自旋标记(ASL)对急性缺血性脑卒中(AIS)患者早期神经功能衰退(END)的预测价值。方法:收集198例AIS患者,根据美国国立卫生研究院脑卒中量表(NIHSS)评分分为END组68例和非END组130例。通过单因素和多因素logistic回归分析评估AIS患者END的相关因素。绘制ROC曲线,评估各相关因素对END的预测价值。结果:2组年龄、性别、NIHSS评分、多个标记后延迟(PLD)(PLD 1.5 s、PLD 2.5 s)的低灌注体积、低灌注体积比(HVR)比较,差异均有统计学意义(均P<0.05)。单因素logistic回归分析显示,年龄、性别、NIHSS评分、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR是AIS患者END的相关因素(均P<0.05)。多因素logistic分析显示,年龄、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR是AIS患者END的独立危险因素(均P<0.05),NIHSS评分为保护因素(P<0.001)。ROC曲线显示,年龄、NIHSS评分、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR预测AIS患者END的AUC分别为0.672、0.702、0.701、0.722、0.902。结论:年龄、NIHSS评分、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR对AIS的END具有预测价值,且HVR的预测价值更高。展开更多
当前基于深度学习的地表覆盖分类方法依赖于大规模且标注精准的训练样本。但受限于成本和技术因素,训练样本不可避免地混入噪声标签,导致分类精度降低。因此,本文提出了融合协同学习和抗噪损失的含噪地表覆盖分类方法。该方法以协同学...当前基于深度学习的地表覆盖分类方法依赖于大规模且标注精准的训练样本。但受限于成本和技术因素,训练样本不可避免地混入噪声标签,导致分类精度降低。因此,本文提出了融合协同学习和抗噪损失的含噪地表覆盖分类方法。该方法以协同学习机制为主体架构,首先利用参数非共享的双支卷积网络分别提取影像初分类特征;然后,基于双支网络的影像分类概率建模干净数据和噪声数据,构建基于信息熵的噪声可信度评价指标;最后,以可信度评价指标代替人工设定权重的方式,提出基于噪声可信度的自适应主动被动损失函数,引导协同学习网络关注噪声样本。试验表明,该方法在公开数据集GID(模拟噪声)和众源数据OSM(实际噪声)上的平均交并比(mean intersection over union,mIOU)分别提升3.85%~23.3%和5.89%~6.73%,说明该方法具有更好的抗噪性能。展开更多
文摘在低标签率场景下,因监督信息极少,现有的半监督方法在获得令人满意的分类结果方面面临挑战。为此提出1种基于自适应正则化补充交叉熵损失的图卷积网络(graph convolutional network based on adaptive regularization complement with cross entropy loss,ARCGCN),从损失函数的设计角度出发,通过引入补充熵与图拉普拉斯正则化项,构建适用于标签稀疏数据集并能够抑制标签噪声影响的半监督图卷积网络,从而有效提升模型预测精度与泛化能力。在6个数据集3Sources、Citeseer、GRAZ02、Out-Scene、Noisy-MNIST和NGs上进行实验,通过与9种基线方法对比,证明了ARCGCN性能的有效性和优越性。结果表明:特别在1%的极低标签率下,ARCGCN相较于所有基线方法,其准确率提升尤为显著,充分验证了其处理标签稀疏数据的能力。
文摘A new concept of labeling called the signed product cordial labeling is introduced and investigated for path graph, cycle graphs, star-K1,n, Bistar-Bn,n, and Some general results on signed product cordial labeling are studied.
文摘X-ray computed tomography(CT),ultrasonography(US)and radionuclide scanning are important clinical methods for evaluating morphology of the kidney.These modalities are also applicable for estimating kidney function with time lapse analysis using proper contrastmedia as may be necessary.In the case of US,it can estimate kidney function based on the measurement of blood flow using the Doppler effect.Formerly,magnetic resonance imaging(MRI)was an inappropriate diagnostic imaging technique for abdominal organs because of their respiratory displacements.However,MRI is now actively used for kidney as well as liver or other parenchymal organs,in tandem with the technological advances.Unlike unenhanced X-ray CT,"conventional"MRI can distinguish the border between cortex and medulla in T1 or T2 weighted images.It was known that the border blurred with decreasing kidney function.Moreover,several other particular imaging methods were introduced in recent years,and these could be called"functional"MRI.In this review,the following are discussed:functional MRI for chronic kidney disease,which include blood oxygenation level-dependent MRI for evaluation of hypoxia,diffusion-weighted imagingfor evaluation of fibrosis,diffusion tensor imaging for evaluation of microstructure,and arterial spin labeling to evaluate the amount of organ perfusion,accompanied with several related articles.The ultimate goal of functional MRI is to provide useful in vivo information repeatedly for daily medical treatment non-invasively.
文摘Arterial spin labeling(ASL) is a magnetic resonance imaging technique for measuring tissue perfusion using a freely diffusible intrinsic tracer.As compared with other perfusion techniques,ASL offers several advantages and is now available for routine clinical practice in many institutions.Its noninvasive nature and ability to quantitatively measure tissue perfusion make ASL ideal for research and clinical studies.Recent technical advances have increased its sensitivity and also extended its potential applications.This review focuses on some basic knowledge of ASL perfusion,emerging techniques and clinical applications in neuroimaging.
文摘目的:评价动脉自旋标记(ASL)对急性缺血性脑卒中(AIS)患者早期神经功能衰退(END)的预测价值。方法:收集198例AIS患者,根据美国国立卫生研究院脑卒中量表(NIHSS)评分分为END组68例和非END组130例。通过单因素和多因素logistic回归分析评估AIS患者END的相关因素。绘制ROC曲线,评估各相关因素对END的预测价值。结果:2组年龄、性别、NIHSS评分、多个标记后延迟(PLD)(PLD 1.5 s、PLD 2.5 s)的低灌注体积、低灌注体积比(HVR)比较,差异均有统计学意义(均P<0.05)。单因素logistic回归分析显示,年龄、性别、NIHSS评分、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR是AIS患者END的相关因素(均P<0.05)。多因素logistic分析显示,年龄、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR是AIS患者END的独立危险因素(均P<0.05),NIHSS评分为保护因素(P<0.001)。ROC曲线显示,年龄、NIHSS评分、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR预测AIS患者END的AUC分别为0.672、0.702、0.701、0.722、0.902。结论:年龄、NIHSS评分、1.5 s PLD低灌注体积、2.5 s PLD低灌注体积、HVR对AIS的END具有预测价值,且HVR的预测价值更高。
文摘当前基于深度学习的地表覆盖分类方法依赖于大规模且标注精准的训练样本。但受限于成本和技术因素,训练样本不可避免地混入噪声标签,导致分类精度降低。因此,本文提出了融合协同学习和抗噪损失的含噪地表覆盖分类方法。该方法以协同学习机制为主体架构,首先利用参数非共享的双支卷积网络分别提取影像初分类特征;然后,基于双支网络的影像分类概率建模干净数据和噪声数据,构建基于信息熵的噪声可信度评价指标;最后,以可信度评价指标代替人工设定权重的方式,提出基于噪声可信度的自适应主动被动损失函数,引导协同学习网络关注噪声样本。试验表明,该方法在公开数据集GID(模拟噪声)和众源数据OSM(实际噪声)上的平均交并比(mean intersection over union,mIOU)分别提升3.85%~23.3%和5.89%~6.73%,说明该方法具有更好的抗噪性能。
