针对眼底图像中存在大量不规则、噪声干扰严重、边界模糊、分割难度较大的细小血管的问题,提出一种基于多方向特征和连通性检测的眼底图像分割方法MDF_Net&CD(Multi-Directional Features neural Network and Connectivity Detecti...针对眼底图像中存在大量不规则、噪声干扰严重、边界模糊、分割难度较大的细小血管的问题,提出一种基于多方向特征和连通性检测的眼底图像分割方法MDF_Net&CD(Multi-Directional Features neural Network and Connectivity Detection)。设计了一个以像素点不同方向特征向量为输入的深度神经网络模型MDF_Net(Multi-Directional Features neural Network),利用MDF_Net对眼底图像进行初步分割;提出连通性检测算法,根据血管的几何特征,对MDF_Net的初步分割结果进一步修订。在公开的眼底图像数据集上,将MDF_Net&CD与近期有代表性的分割方法进行实验对比,结果表明MDF_Net&CD各项评估指标均衡,敏感度,F1值和准确率优于其他方法。该方法能有效捕捉像素点的细节特征,对不规则、噪声干扰严重、边界模糊的细小血管有较好分割效果。展开更多
Due to the increasing number ot diabetic patients, the number of people affected by diabetic retinopathy isexpected to increase. Diabetic retinopathy is a complication of diabetes and the most serious frequent eye dis...Due to the increasing number ot diabetic patients, the number of people affected by diabetic retinopathy isexpected to increase. Diabetic retinopathy is a complication of diabetes and the most serious frequent eye disease in the world. Large-scale retinal screening for diabetic patients is necessary to prevent visual loss and blindness. The analysis of digital retinal images, obtained by the fundus camera, is viewed as a feasible approach because retinal blood vessels have been shown to change in diameter, branching angles, or tortuosity as a result of diabetic retinopathy. The morphological change can help identify the different stages of diabetic retinopathy. In addition, the acquisition of retinal image is nonintrusive and low cost. Automatic segmentation of the retinal blood vessel is a prerequisite for this analysis.~3 This article presents a method to detect blood vessel based on sobel operators.4 Small and fast computation is the outstanding merit of this method.展开更多
文摘针对眼底图像中存在大量不规则、噪声干扰严重、边界模糊、分割难度较大的细小血管的问题,提出一种基于多方向特征和连通性检测的眼底图像分割方法MDF_Net&CD(Multi-Directional Features neural Network and Connectivity Detection)。设计了一个以像素点不同方向特征向量为输入的深度神经网络模型MDF_Net(Multi-Directional Features neural Network),利用MDF_Net对眼底图像进行初步分割;提出连通性检测算法,根据血管的几何特征,对MDF_Net的初步分割结果进一步修订。在公开的眼底图像数据集上,将MDF_Net&CD与近期有代表性的分割方法进行实验对比,结果表明MDF_Net&CD各项评估指标均衡,敏感度,F1值和准确率优于其他方法。该方法能有效捕捉像素点的细节特征,对不规则、噪声干扰严重、边界模糊的细小血管有较好分割效果。
文摘Due to the increasing number ot diabetic patients, the number of people affected by diabetic retinopathy isexpected to increase. Diabetic retinopathy is a complication of diabetes and the most serious frequent eye disease in the world. Large-scale retinal screening for diabetic patients is necessary to prevent visual loss and blindness. The analysis of digital retinal images, obtained by the fundus camera, is viewed as a feasible approach because retinal blood vessels have been shown to change in diameter, branching angles, or tortuosity as a result of diabetic retinopathy. The morphological change can help identify the different stages of diabetic retinopathy. In addition, the acquisition of retinal image is nonintrusive and low cost. Automatic segmentation of the retinal blood vessel is a prerequisite for this analysis.~3 This article presents a method to detect blood vessel based on sobel operators.4 Small and fast computation is the outstanding merit of this method.
