提出了一种基于多尺度特征融合的全卷积神经网络的视网膜血管分割方法,无需手工设计特征和后处理过程。利用跳跃连接构建编码器-解码器结构全卷积神经网络,将高层语义信息和低层特征信息进行融合;利用残差块进一步学习细节和纹理特征;...提出了一种基于多尺度特征融合的全卷积神经网络的视网膜血管分割方法,无需手工设计特征和后处理过程。利用跳跃连接构建编码器-解码器结构全卷积神经网络,将高层语义信息和低层特征信息进行融合;利用残差块进一步学习细节和纹理特征;利用不同空洞率的空洞卷积构建多尺度空间金字塔池化结构,进一步扩大感受野,充分结合图像上下文信息;采用类别平衡损失函数解决正负样本不均衡问题。实验结果表明,在DRIVE(Digital Retinal Images for Vessel Extraction)和STARE (Structured Analysis of the Retina)数据集上的准确率分别为95.46%和96.84%,敏感性分别为80.53%和82.99%,特异性分别为97.67%和97.94%,受试者工作特征(ROC)曲线下的面积分别为97.71%和98.17%。所提方法相较于其他方法性能更优。展开更多
针对视网膜血管形态结构和尺度信息复杂多变的特点,提出一种自适应血管形态结构和尺度信息的U型视网膜血管分割算法。首先采用二维K-L(Karhunen-Loeve)变换(即霍特林变换)综合分析彩色图像三通道的频带信息,从而得到视网膜灰度图像以及...针对视网膜血管形态结构和尺度信息复杂多变的特点,提出一种自适应血管形态结构和尺度信息的U型视网膜血管分割算法。首先采用二维K-L(Karhunen-Loeve)变换(即霍特林变换)综合分析彩色图像三通道的频带信息,从而得到视网膜灰度图像以及多尺度形态学滤波增强血管与背景的对比度信息。然后将预处理图像经U型分割模型对图像进行端对端训练,并利用局部信息熵采样进行数据增强。该网络编码部分的密集可变形卷积结构根据上下特征层信息有效地捕捉图像中多种尺度信息和形状结构,底部金字塔型的多尺度空洞卷积扩大局部感受野,同时解码阶段带有Attention机制的反卷积网络将底层与高层特征映射有效结合,解决权重分散和图像纹理损失的问题。最后通过SoftMax激活函数得到最终的分割结果。在DRIVE(Digital Retinal Images for Vessel Extraction)与STARE(Structured Analysis of the Retina)数据集上对该算法进行了仿真,准确率分别达到97.48%与96.83%,特异性分别达到98.83%与97.75%,总体性能优于现有算法。展开更多
当前主流的眼底视网膜血管分割方法存在细微血管细粒度特征很难采集和细节容易丢失的问题。为解决这一问题,设计了一种改进U-Net模型算法,该算法将U-Net上下采样中的原始卷积层改为二次循环残差卷积层,提升了特征的使用效率;在解码部分...当前主流的眼底视网膜血管分割方法存在细微血管细粒度特征很难采集和细节容易丢失的问题。为解决这一问题,设计了一种改进U-Net模型算法,该算法将U-Net上下采样中的原始卷积层改为二次循环残差卷积层,提升了特征的使用效率;在解码部分引入多通道注意力模型,改善了低对比度下细小血管的分割效果。该算法在DRIVE(Digital Retinal Images for Vessel Extraction)和STARE(Structured Analysis of the Retina)两个数据库的准确率分别为96.89%和97.96%,敏感度分别为80.28%和82.27%,AUC(Area Under Curve)性能分别为98.41%和98.65%,较现有的先进算法有一定的提升。本文所提算法能有效提高眼底图像细微血管分割准确率。展开更多
Objective To investigate the association of retinal vascular calibers with hyperuricemia in a middle‐aged and elderly population. Methods A cross‐sectional design was applied in this study and 869 participants aged ...Objective To investigate the association of retinal vascular calibers with hyperuricemia in a middle‐aged and elderly population. Methods A cross‐sectional design was applied in this study and 869 participants aged ≥40 years from a high‐risk group for diabetes were recruited. All participants received the anthropometrical measurements and laboratory tests. Retinal arteriolar and venular caliber of the participants were measured with a semi‐automated system. Hyperuricemia was defined as a serum uric acid level 420 μmol/L in men and 360 μmol/L in women. Linear regression models were used to assess the association of hyperuricemia with retinal vascular calibers. These models were additionally adjusted for age, central obesity, hypertension, dyslipidemia, weekly activity, smoking status, and education. Results Among the 869 participants, 133 （15.3%） suffered from hyperuricemia. The crude mean serum uric acid level was 312.3 μmol/L （Standard Deviation 79.5）; mean concentration was 355.0 μmol/L （SD 75.5） in male participants, and 288.0 μmol/L （SD 71.1） in female participants （age‐adjusted difference 58.1 μmol/L, 95% Confidence Internal 48.5, 67.6）. After adjusting for additional covariates, male participants with hyperuricemia had 3.77 μm （95% CI ‐0.46, 8.00） smaller arteriolar caliber and 6.20 μm （95% CI 0.36, 12.04） larger venule than those without hyperuricemia; the corresponding numbers among female participants were 1.57 μm （95% CI ‐1.07, 4.21） for retinal arteriolar caliber and 2.28 μm （95% CI ‐1.72, 6.27） for retinal venular caliber. Conclusion Hyperuricemia was associated with smaller retinal arteriolar caliber and larger venular caliber mainly in male participants in this study.展开更多
文摘提出了一种基于多尺度特征融合的全卷积神经网络的视网膜血管分割方法,无需手工设计特征和后处理过程。利用跳跃连接构建编码器-解码器结构全卷积神经网络,将高层语义信息和低层特征信息进行融合;利用残差块进一步学习细节和纹理特征;利用不同空洞率的空洞卷积构建多尺度空间金字塔池化结构,进一步扩大感受野,充分结合图像上下文信息;采用类别平衡损失函数解决正负样本不均衡问题。实验结果表明,在DRIVE(Digital Retinal Images for Vessel Extraction)和STARE (Structured Analysis of the Retina)数据集上的准确率分别为95.46%和96.84%,敏感性分别为80.53%和82.99%,特异性分别为97.67%和97.94%,受试者工作特征(ROC)曲线下的面积分别为97.71%和98.17%。所提方法相较于其他方法性能更优。
文摘针对视网膜血管形态结构和尺度信息复杂多变的特点,提出一种自适应血管形态结构和尺度信息的U型视网膜血管分割算法。首先采用二维K-L(Karhunen-Loeve)变换(即霍特林变换)综合分析彩色图像三通道的频带信息,从而得到视网膜灰度图像以及多尺度形态学滤波增强血管与背景的对比度信息。然后将预处理图像经U型分割模型对图像进行端对端训练,并利用局部信息熵采样进行数据增强。该网络编码部分的密集可变形卷积结构根据上下特征层信息有效地捕捉图像中多种尺度信息和形状结构,底部金字塔型的多尺度空洞卷积扩大局部感受野,同时解码阶段带有Attention机制的反卷积网络将底层与高层特征映射有效结合,解决权重分散和图像纹理损失的问题。最后通过SoftMax激活函数得到最终的分割结果。在DRIVE(Digital Retinal Images for Vessel Extraction)与STARE(Structured Analysis of the Retina)数据集上对该算法进行了仿真,准确率分别达到97.48%与96.83%,特异性分别达到98.83%与97.75%,总体性能优于现有算法。
文摘当前主流的眼底视网膜血管分割方法存在细微血管细粒度特征很难采集和细节容易丢失的问题。为解决这一问题,设计了一种改进U-Net模型算法,该算法将U-Net上下采样中的原始卷积层改为二次循环残差卷积层,提升了特征的使用效率;在解码部分引入多通道注意力模型,改善了低对比度下细小血管的分割效果。该算法在DRIVE(Digital Retinal Images for Vessel Extraction)和STARE(Structured Analysis of the Retina)两个数据库的准确率分别为96.89%和97.96%,敏感度分别为80.28%和82.27%,AUC(Area Under Curve)性能分别为98.41%和98.65%,较现有的先进算法有一定的提升。本文所提算法能有效提高眼底图像细微血管分割准确率。
基金supported by Science and Technology Commission of Shanghai Municipality (STCSM) and the Key Project of Health Bureau of Shanghai (Grant 04dz19501‐1 and 08GWZX0203 to Xin GAO)
文摘Objective To investigate the association of retinal vascular calibers with hyperuricemia in a middle‐aged and elderly population. Methods A cross‐sectional design was applied in this study and 869 participants aged ≥40 years from a high‐risk group for diabetes were recruited. All participants received the anthropometrical measurements and laboratory tests. Retinal arteriolar and venular caliber of the participants were measured with a semi‐automated system. Hyperuricemia was defined as a serum uric acid level 420 μmol/L in men and 360 μmol/L in women. Linear regression models were used to assess the association of hyperuricemia with retinal vascular calibers. These models were additionally adjusted for age, central obesity, hypertension, dyslipidemia, weekly activity, smoking status, and education. Results Among the 869 participants, 133 （15.3%） suffered from hyperuricemia. The crude mean serum uric acid level was 312.3 μmol/L （Standard Deviation 79.5）; mean concentration was 355.0 μmol/L （SD 75.5） in male participants, and 288.0 μmol/L （SD 71.1） in female participants （age‐adjusted difference 58.1 μmol/L, 95% Confidence Internal 48.5, 67.6）. After adjusting for additional covariates, male participants with hyperuricemia had 3.77 μm （95% CI ‐0.46, 8.00） smaller arteriolar caliber and 6.20 μm （95% CI 0.36, 12.04） larger venule than those without hyperuricemia; the corresponding numbers among female participants were 1.57 μm （95% CI ‐1.07, 4.21） for retinal arteriolar caliber and 2.28 μm （95% CI ‐1.72, 6.27） for retinal venular caliber. Conclusion Hyperuricemia was associated with smaller retinal arteriolar caliber and larger venular caliber mainly in male participants in this study.
