基于多尺度残差卷积神经网络的视杯视盘联合分割

青光眼是不可逆性失明的首要原因,早期症状不明显,容易被忽视,因此青光眼早期筛查尤为重要。杯盘比是临床上用于青光眼筛查的重要指标,所以精准分割视杯视盘是计算杯盘比的关键。本文提出了一种基于全卷积多尺度残差神经网络的视杯视盘分割方法。首先,对眼底图像进行对比度增强,并引入极坐标变换。随后,构造W-Net作为主体网络,用带残差多尺度全卷积模块来替代标准卷积单元,输入端口加入图像金字塔来构造多尺度输入,侧输出层作为早期的分类器生成局部预测输出。最后,提出了一种新的多标签损失函数来指导网络分割。实验采用REFUGE数据集验证,最终视杯、视盘分割的平均交并比分别为0.904 0、0.955 3,重叠误差分别为0.178 0、0.066 5。结果表明,该方法不仅实现了视杯视盘联合分割,而且有效提高了其分割精度。该方法将有助于大规模青光眼早期筛查的推广。     

基于线性化注意力和双重注意力的视杯盘分割模型

青光眼是致盲性眼病之一，视杯盘比是筛查青光眼的主要依据，因此准确分割视杯盘具有重要意义。本文提出一种基于线性化注意力和双重注意力的视杯盘分割模型。首先，根据视盘特性定位裁剪感兴趣区域。其次，引入线性化注意力的残差网络-34(ResNet-34)作为特征提取网络。最后，通过线性化注意力的输出特征生成通道和空间双重注意力权重，用于校准解码器输出特征获取视杯盘分割图像。实验结果表明，所提模型在视神经头分割的视网膜图像（DRISHTI-GS）数据集中，视盘、视杯交并比分别为0.962 3、0.856 4；用于视神经评估的开放式视网膜图像-V3(RIM-ONE-V3)数据集中，视盘、视杯交并比分别为0.956 3、0.784 4。所提模型优于对比算法，在青光眼的早期筛查中具有一定的医学价值。此外，本文利用知识蒸馏技术生成两种规模更小的模型，有利于将模型应用于嵌入式设备。     

基于多层Mumford-Shah向量模型的彩色视乳头图像杯盘重建、分割与度量

基于多层Mumford-Shah向量值图像分割、去噪与重建模型(HMSMv)和光滑样条曲线拟合技术,提出了一种用于计算机辅助青光眼诊断的视乳头图像视杯和视盘重建、分割与度量的新方法。首先,采用HMSMv分割和重建视杯和视盘;然后,基于重建的视乳头图像,结合青光眼视乳头图像杯、盘的先验知识,提取视杯和视盘特征矩形和边缘特征点;最后,采用光滑样条曲线拟合技术,重建被血管遮挡的视杯和视盘部分边缘,并计算杯盘比等病理特征参数值。不同青光眼病人的视乳头图像杯盘重建、分割与度量实验结果表明,该方法能克服噪声污染、血管遮挡、光照不均匀、对比度小、个体间差异大等视网膜图像分割中固有的困难,并有效重建、分割与度量青光眼彩色视乳头图像中的视杯和视盘。     

一种基于SFS的视乳头三维形貌重建方法

视乳头形貌的确定有助于眼底类病变的准确诊断,例如视杯和视盘尺寸的相对大小是青光眼疾病病程诊断的重要依据。通常视盘和视杯的大小是通过眼底照片分离视乳头区域并进行测量,这种方法对视乳头的三维形变缺乏全面的描述。为了更加方便准确地获取视杯视盘大小的比值和视乳头的三维形貌,采用明暗恢复形状的计算机视觉方法,通过眼底图像来重建眼底结构的三维形貌。具体步骤包括:通过眼底照相机,得到二维眼底数字图像;采用图像分割方法,得到视杯和视盘区域;估计视盘视杯图像区域的光源照射方向;利用Tsai线性化方法,根据二维数字眼底图像的明暗信息,重建眼底三维结构形貌图。对初步实验结果中视杯和视盘的比计算可得,青光眼患者的杯盘比的比值大于0.6,与临床上对青光眼的判断相吻合。此方法提供新的眼底形貌变形可视化方法,能够清晰、准确地重建出眼底视盘和视杯的三维结构形貌图;可以对视盘视杯几何尺寸和结构进行更全面的描述,为眼底疾病方便和准确地诊断提供新的线索。     

结合上下文和注意力机制改进的视盘分割模型

青光眼为多发性眼底疾病,是致盲的主要原因之一。眼底图像来源广,质量参差不齐,且视盘区域具有多尺度性特征,融合上下文信息有利于准确分割多尺度视盘边界。以U-Net为基础,结合上下文信息和卷积注意力模块（CBAM）,提出了一种改进的视盘分割模型,包括：（1）使用实例-批处理归一化（IBN）模块与注意力机制改进主干网络ResNet34,提升分割模型的泛化性和图像通道特征的提取能力;（2）提出一种多层次上下文信息提取（MCE）模块处理主干网络输出的特征,融合上下文信息增强分割模型对视盘边缘特征的提取能力;（3）使用Transformer机制替换U-Net中的跳跃连接和上采样,进一步提高视盘多尺度特征和图像通道特征的提取能力。将改进的分割模型与U-Net、U-Net++、DeeplabV3+、FCN和PSPNet分割模型进行视盘分割精度比较,结果表明提出的分割模型具有更好的分割效果,Dice、MIoU、MPA和FPS指标分别为98.18%、96.45%、98.11%和17.56 Img/s。该研究成果可为青光眼的早期诊断提供技术支撑。     

一种眼底视盘图像的自动分割方法

本文研究了基于计算机图像处理技术的眼底视盘图像的自动分割方法。首先获取视盘图像的红基色图像,然后根据红基色图像中视盘的明显的边沿特征实现视盘的自动分割。一幅具体的眼底图像实验证明了本方法的可行性。     

视觉诱发电位在青光眼诊断及研究中的应用

青光眼是由于眼内压升高而引起视乳头及视神经纤维层损害和相应视功能丧失的一种严重致盲性疾病。特别是原发性开角型青光眼,因其发病及进展隐匿,不易及早发现和进行病情监测而对人类健康造成更大成协。近代科学技术的发展已经证明传统的诊断方法远不能对青光眼做出早期诊断,如眼压升高和视乳头杯盘比的扩大在青光眼病人和正常人群间存在着较大的交叉,而在获得确凿的青光眼性视野缺损时则已有大约40％以上的视网膜神经纤维遭到损害;同样,一些传统的激发试验,如饮水或激素试验也是不可靠的。目前许多眼科专家     

基于三重注意力的脑肿瘤图像分割网络

脑肿瘤图像分割问题是脑肿瘤临床诊断和治疗脑肿瘤疾病计算机辅助诊断的基础.针对脑肿瘤MRI图像分割网络深度过深和局部与全局特征信息联系匮乏导致图像分割精度降低等问题,提出一种基于三重注意力的脑肿瘤图像分割网络.首先,借鉴残差结构,将原始图像分割网络结构的编码层和解码层中的卷积模块替换为深度残差模块,解决网络加深带来的梯度...     

基于多尺度编码-解码网络的皮肤病变图像分割

针对皮肤病变图像分割在医疗诊断中的作用,提出一种基于多尺度编码-解码网络的皮肤病变图像分割算法。该算法继承了SegNet网络结构的训练速度快、训练模型存储小等特点,采用多尺度输入的方式增强了网络对皮肤病变图像的充分学习。此外,在编码网络中的pool2层输出一个二进制双线性插值的中间预测特征图到解码层的最后一层卷积块进行级联输入提高最终的分割精度。实验结果表明,采用多尺度编码-解码网络对皮肤病变图像分割具有极好的效果,在其他医学图像分割方面也能进行广泛应用。     

基于边缘监督的肝部超声图像包膜分割网络

肝纤维化、肝硬化的早期发现对临床治疗和预后评估具有重要意义。而肝包膜的形态和纹理特征是计算机辅助肝硬化诊断的重要依据。本文提出一种基于边缘监督的肝部超声图像包膜分割网络。该网络以常用的分割模型UNet为基础，引入空洞卷积，扩大感受野；同时，添加了边缘监督模块，从而将特征学习主要聚焦在图像梯度较大的部分；此外，还设计了混合加权损失函数，来缓解肝包膜部分与其他区域之间的极度不平衡情况。实验结果表明，本文提出的ES-UNet网络结构平均Dice系数相比原始UNet提高了0.171 5，平均交并比（MIoU）提高了0.021 5，其他指标也有较明显的提高，可见，本文算法的各个组件对模型分割性能的优化都有一定的贡献，改进后的模型可以实现肝包膜的精确分割。     

Automated determination of cup-to-disc ratio for classification of glaucomatous and normal eyes on stereo retinal fundus images

Early diagnosis of glaucoma, which is the second leading cause of blindness in the world, can halt or slow the progression of the disease. We propose an automated method for analyzing the optic disc and measuring the cup-to-disc ratio (CDR) on stereo retinal fundus images to improve ophthalmologists' diagnostic efficiency and potentially reduce the variation on the CDR measurement. The method was developed using 80 retinal fundus image pairs, including 25 glaucomatous, and 55 nonglaucomatous eyes, obtained at our institution. A disc region was segmented using the active contour method with the brightness and edge information. The segmentation of a cup region was performed using a depth map of the optic disc, which was reconstructed on the basis of the stereo disparity. The CDRs were measured and compared with those determined using the manual segmentation results by an expert ophthalmologist. The method was applied to a new database which consisted of 98 stereo image pairs including 60 and 30 pairs with and without signs of glaucoma, respectively. Using the CDRs, an area under the receiver operating characteristic curve of 0.90 was obtained for classification of the glaucomatous and nonglaucomatous eyes. The result indicates potential usefulness of the automated determination of CDRs for the diagnosis of glaucoma.     

An Efficient Hierarchical Optic Disc and Cup Segmentation Network Combined with Multi-task Learning and Adversarial Learning

Journal of Digital Imaging - Automatic and accurate segmentation of optic disc (OD) and optic cup (OC) in fundus images is a fundamental task in computer-aided ocular pathologies diagnosis. The...     

眼底数码图像处理关键技术研究

目的目前医生一般通过检眼镜等设备查看眼病患者眼底结构特征诊断疾病,工序烦琐、存在误差并受主观因素的影响,诊断效率低。因此有必要开发一款眼底数码图像处理系统配合医生诊断。方法对原始图像进行增强对比度和空域滤波的预处理操作、二值化处理、形态学处理及边缘提取和骨架提取,进而得到目标特征区域。结果本文初步研究眼底数码图像视杯、视盘和血管轮廓特征的自动提取关键技术,通过MATLAB仿真软件对特征提取关键技术进行总结和验证,归纳出提取不同眼底特征时的应对策略。结论利用数字图像处理技术可以自动识别眼底数码图像中的不同特征区域,建立眼底数码图像处理系统用于辅助医生诊断是可行的。     

双线性非局部特征结合中继监督网络用于视网膜血管分割

对眼底图像中的视网膜血管精准分割是检测多种疾病的关键技术,在相关疾病自动筛查系统中发挥着重要的作用。针对现存方法追求分割精度时忽略对算法复杂性的考虑,导致在资源受限的医疗设备上部署困难的问题。本文通过进一步合理减少卷积层的特征通道数量来轻量化分割网络并提出了BNIS-Net。该网络采用多尺度图像作为输入融合到编码过程中,使得不同感受野之间建立良好的联系,并提出一种双线性非局部模块来增强相关上下文信息的捕捉能力。最后,在解码过程中采用中继监督的策略,为解码部分各级输出提供监督来约束网络的学习,这样可以有效改善收敛行为使浅层部分得到充分训练。BNIS-Net以0.41 M的参数量在DRIVE、STARE和CHASE_DB1 3个公开数据集上分别取得了81.02%、81.07%、78.15%的DSC值和0.983 3、0.986 1、0.985 9的AUC值。通过大量对比实验和消融研究证明,该方法能够更好地分割血管的边缘细节。     

The in vivo 3D Optic Nerve Head Modeling Based on Human Multimodality Images

Introduction: The human optic nerve head(ONH) is vulnerable to the damage in glaucomatous high intraocular pressure (IOP). In order to analyze the human ONH head stress and deformation in high IOP, an in vivo three-dimensional (3D) ONH model was reconstructed by optical coherence tomography (OCT) images and magnetic resonance imaging (MRI) images. Materials and Methods: A human eye was scanned by MRI and OCT in serial imaging protocol. The sclera and ONH were segmented from the images, and 3D models were reconstructed by multimodality image registration. Through the morphological segmentation, part of lamina cribrosa (LC) was acquired and reconstructed in combination with the ONH and sclera. Results: The models of ONH and sclera were got, the part of LC was included in the model. In the analysis of FEM, the ONH was compressed and the cup/disk ratio was changed obviously in high glaucomatous IOP. Discussion: This study described a method to build a 3D in vivo ONH model by image processing. It can be used in biomechanical analysis, and provide the stress state of ONH for the research about the fundus damage of glaucoma.     

Optic disc detection in color fundus images using ant colony optimization

Diabetic retinopathy has been revealed as the most common cause of blindness among people of working age in developed countries. However, loss of vision could be prevented by an early detection of the disease and, therefore, by a regular screening program to detect retinopathy. Due to its characteristics, the digital color fundus photographs have been the easiest way to analyze the eye fundus. An important prerequisite for automation is the segmentation of the main anatomical features in the image, particularly the optic disc. Currently, there are many works reported in the literature with the purpose of detecting and segmenting this anatomical structure. Though, none of them performs as needed, especially when dealing with images presenting pathologies and a great variability. Ant colony optimization (ACO) is an optimization algorithm inspired by the foraging behavior of some ant species that has been applied in image processing with different purposes. In this paper, this algorithm preceded by anisotropic diffusion is used for optic disc detection in color fundus images. Experimental results demonstrate the good performance of the proposed approach as the optic disc was detected in most of all the images used, even in the images with great variability.     

Role of cerebrospinal fluid in glaucoma: Pressure and beyond

Glaucoma is the second leading cause of blindness worldwide. Primary open-angle glaucoma (POAG) is characterized by optic disc cupping and visual field impairment. Though the elevated intraocular pressure (IOP) is thought to be the major risk factor for POAG, about 50% of the POAG patients have normal IOP, called ‘normal-tension’ glaucoma. Besides, many POAG patients still experience visual field loss and/or optic disc cupping even though the IOP has been well controlled. The mechanisms underlying the pathogenesis of POAG remain unclear. Extensive studies have shed lights on the mechanisms that may be involved in the etiopathology and/or the optic neuropathic manifestations of POAG. In this article, we noticed that the changes in the cerebrospinal fluid, particularly that existing in the subarachnoid space of the optic nerve, appear to be actively involved in the pathogenesis of POAG.     

基于改进U型网络的眼底图像血管分割

眼底图像血管分割问题是眼科及其他相关疾病计算机辅助诊断的基础。通过分割和分析眼底图像中的血管结构,可以对糖尿病视网膜病变、高血压和动脉硬化等疾病进行早期诊断和监测。针对目前已有血管分割算法存在准确率不高和灵敏度较低的问题,基于深度学习基本理论,提出一种改进U型网络的眼底图像血管分割算法。首先,通过减少传统U型网络下采样和上采样操作次数,解决眼底图像数据较少的问题;其次,通过将传统卷积层串行连接方式改为残差映射相叠加的方式,提高特征的使用效率;最后,在卷积层之间加入批量归一化和PReLU激活函数对网络进行优化,使网络性能得到进一步的提升。在DRIVE和CHASE_DB1这两个公开的眼底数据库上进行实验,每个数据库随机抽取160 000个图像块送入改进的网络中进行训练和测试,可以得到该算法在两个数据库上的灵敏度、准确率和AUC(ROC曲线下的面积)值,相比已有算法的最好结果平均分别提高2.47%、0.21%和0.35%。所提出的算法可改善眼底图像细小血管分割准确率不高及灵敏度较低的问题,能够较好地分割出低对比度的微细血管。