基于U-Net神经网络眼底图像的视盘分割研究

眼底图像中的视盘在青光眼筛查和诊断中起着重要作用。因此,从眼底图像中对视盘进行准确、快速地定位与分割具有重要意义。在过去,研究者们已经进行了对视盘的深入研究,但如何提高定位准确率和分割精度仍是视盘分割的一大难题。对此本文提出一种采用深度学习结构U-Net的视网膜视盘自动分割的方法,该方法结合机器学习,通过深度网络提取输入图像的视盘特征,从而得出相应的分割结果图。相对于传统的视盘分割方法,本文的U-Net神经网络能够有效学习有利于分割视盘的特征,从而提高分割的精确度,而且分割耗时更短。     

心血管造影图像的分割与诊断

针对心血管造影图像,提出了一种新的血管分割方法,并实现了血管狭窄位置的自动诊断。血管分割中提出了基于Hessian矩阵多尺度增强的新血管函数,在增强的特征图上采用了种子点的自动化选取和双阶段区域增长的分割方法提取了血管轮廓。在此基础上进行了血管骨架提取,骨架点搜索和直径测量,实现了全自动的狭窄诊断。实验结果表明,新的血管分割方法能够提取出较精准的主分支轮廓,对细小血管也有良好的增强和分割效果,血管狭窄的自动诊断,计算速度快,诊断结果较准确,能够对医生的最终判断提供辅助参考和量化依据。     

基于Hessian矩阵的视网膜血管中心线提取

为实现视网膜血管在临床诊断中的重要作用,提出了一种基于Hessian矩阵的视网膜血管提取方法,该方法通过图像预处理增强血管信息,利用血管的微分几何特性,采用离散高斯核对眼底图像进行卷积,结合Hessian矩阵计算血管方向,通过连接算法得到视网膜血管的分布情况.实验结果表明,该方法提取血管中心线的精度可达亚像素级,对不同眼底照相机拍摄的眼底图像可根据血管宽度进行多尺度快速分割.     

彩色眼底图像视盘自动定位与分割

针对彩色眼底图像视盘定位时图像边缘高亮环对定位准确率的影响,提出了一种有效的图像预处理方法。针对已有的视盘分割算法中存在的问题,提出了一种结合形态学、椭圆拟合及梯度矢量流(GVF)Snake模型的分割算法。提出的预处理方法首先利用最小二乘法拟合出眼底图像的边界,然后裁剪掉边界的一部分高亮像素点,最后进行视盘定位。视盘分割算法则首先进行血管擦除,然后用椭圆拟合提取初始轮廓,最后使用GVF Snake精确调整视盘边界。用提出的方法对Messidor眼底图像数据库1 200幅图像上进行了实验,结果显示:视盘定位准确率由原来没经过预处理的95.4%提升到了98.7%;视盘分割错误率与当前已知最好的算法相比由12.5%降低到了9.39%。结果表明:提出的眼底图像视盘自动定位与分割方法准确率高、实用性强,可以用于眼科疾病的计算机辅助诊断。     

注意力感知的多尺度语义视杯盘分割

为了解决编码器-解码器网络结构在目标提取中抑制无关语义、跨越语义鸿沟等问题，以获取更高精度，采用UNet作为提取特征的主干网络；为了减轻浅层特征与深层特征语义的差异，设计一种融合注意力感知的多尺度语义池化模块（Channel-Spatial-Pyramid, CSP），替代早期层中的跳跃链接。CSP模块从空间与通道两个层面强调更有意义的语义信息，通过4个不同池化核的并行分支提取不同尺度特征，聚合所有分支结果与后期层特征拼接。实验结果表明，CSP-Net在彩色眼底图像视盘分割中的Dice指数可达99.6%，视杯分割结果的Dice指数为92.1%，相比现有算法均有提高。所提出的CSP-Net对于眼底图像中的微小目标提取的有效性及抗干扰性较高，可为青光眼筛查与诊断临床提借鉴。     

基于Hessian矩阵线状滤波的眼底图像血管分割

针对眼底图像血管分割问题,本文研究了一种基于Hessian矩阵多尺度线状滤波的血管分割方法。首先,采用基于Hessian矩阵的多尺度线状滤波增强血管区域,然后对增强后图像采用最大类间方差阈值法进行阈值分割,最后得到血管的二值化分割结果。本文采用了DRIVE数据库眼底图像进行实验验证,实验表明本文方法能够自动地得到较完整和准确的眼底图像血管分割结果。     

彩色视网膜血管图像的自动分割算法

为了保证眼底测量结果的准确性、客观性、可重复性以及实用性，提出了彩色眼底图像自动分割与定量分析的算法。具体步骤如下：首先对彩色视网膜血管图像进行网格划分，其次对包含重要血管信息的网格区域实现Otsu阈值分割，在此基础上对其它相邻网格进行区域生长算法分割，最后由计算机统一处理得到视网膜血管的网络径线。实验结果表明：该算法提取的血管网络径线连续性较好，血管中心线定位准确，抗干扰能力较强，处理速度较快，具有较高的临床应用价值。     

基于U型卷积网络的视网膜血管分割方法

视网膜血管的结构信息对眼科疾病的诊断具有重要的指导意义,对视网膜血管图像进行高效正确的分割成为临床的迫切需求。为此,提出了一种U型卷积网络,实现了更高效的自动化视网膜血管分割。骨干网络基于经典的编解码架构,编码器采用预训练的残差模块充分提取每一层的特征,解码器通过转置卷积逐层进行上采样,增加了特征的复用性。网络在中间层引入ASPP(Atrous Spatial Pyramid Pooling)模块,提取不同尺度的视网膜血管特征。为了在类内预测上保持一致,在跳级层利用通道注意力模块对特征进行自适应细化,融合了不同层次的特征。在DRIVE数据集上的实验结果表明,与其他相关算法性能相比,该算法的敏感性、特异性、准确率均最高,模型泛化能力好,大大提高了视网膜血管分割的准确性。     

眼底图像动静脉宽度差变化率测量方法

眼底图像中动静脉血管的宽度变化能够体现患者糖尿病的病程情况,人工确定血管宽度及其差值耗时费力且需要丰富经验。本文针对辽宁何氏眼科医院眼病筛查系统中受检者的眼底图像提出了动静脉宽度差变化率的概念,指出了其与糖尿病诊断之间的关系,并选取RGB、LAB、YCb Cr、Gaussian 4个颜色空间中不同的通道分量,定义了新的基于血管中心线像素和血管像素的特征向量,采用线性判别分析(LDA)分类器完成了对动静脉的全自动分类,继而实现了对动静脉血管宽度测量及相应差值变化率的计算。实验结果表明,与其他算法相比,本文方法在血管分割、动静脉分类、血管宽度及差值计算等方面都较准确,其诊断结果与临床诊断基本一致,具有一定的临床应用价值。     

基于活动轮廓模型和Contourlet多分辨率分析分割血管内超声图像

血管内超声斑块图像的分割对动脉粥样硬化疾病的诊断有重要价值。针对传统分割方法初始化和鲁棒性两个问题,提出一种基于活动轮廓模型和Contourlet多分辨率分析分割血管内超声斑块图像的新方法。该方法运用Contourlet变换将原图像分解为多分辨率低通分量和多分辨率带通分量方向性子带。对低通分量进行模板匹配,确定血管内腔边界和中外膜边界的初始轮廓;对带通分量方向性子带进行扩散滤波,抑制噪声的同时尽可能保留有用边缘,并结合边界矢量场使轮廓演化得到最终分割结果,从而提高了分割算法的鲁棒性。对100幅仿真图像和120幅实际图像的分割结果表明,该方法能自动、精确地提取血管的两条边界。相对于传统活动轮廓模型法,该方法分割实际图像的平均距离误差提高了3.04像素,面积差异百分比提高了6.30%。     

Microscopic retinal blood vessels detection and segmentation using support vector machine and K-nearest neighbors

The retina is the deepest layer of texture covering the rear of the eye, recorded by fundus images. Vessel detection and segmentation are useful in disease diagnosis. The retina's blood vessels could help diagnose maladies such as glaucoma, diabetic retinopathy, and blood pressure. A mix of supervised and unsupervised strategies exists for the detection and segmentation of blood vessels images. The tree structure of retinal blood vessels, their random area, and different thickness have caused vessel detection difficulties at machine learning calculations. Since the green band of retinal images conveys more information about the vessels, they are utilized for microscopic vessels detection. The current research proposes an administered calculation for segmentation of retinal vessels, where two upgrading stages depending on filtering and comparative histogram were applied after pre-processing and image quality improvement. At that point, statistical features of vessel tracking, maximum curvature and curvelet coefficient are extracted for each pixel. The extracted features are classified by support vector machine and the k-nearest neighbors. The morphological operators then enhance the classified image at the final stage to segment with higher accuracy. The dice coefficient is utilized for the evaluation of the proposed method. The proposed approach is concluded to be better than different strategies with a normal of 92%.     

结合多尺度圆形滤波与MS聚类 的疑似结节分割

肺部CT图像中疑似病灶感兴趣区域(ROI)的准确分割是肺部计算机辅助检测/诊断(CAD)的重要环节。本文提出结合Hessian矩阵滤波的均值漂移聚类肺部疑似病灶ROI区域分割算法。对原图像进行多尺度Hessian矩阵圆形滤波,图像中圆形的疑似结节病灶区域得到滤波增强、直线形的气管/血管区域得到抑制,将Hessian矩阵滤波后的形状特征、灰度、空间位置3种信息引入特征空间,将均值漂移聚类的核函数分解为3种特征信息所分别对应的核函数乘积形式,最后采用自适应计算带宽的方法确定每个待分割疑似区域的带宽进行均值漂移聚类分割。对来自LIDC等127个包含不同类型肺结节的病例进行实验,实验结果表明引入Hessian矩阵圆形滤波信息的均值漂移聚类能够分割出与血管或气管相连或者交叉的结节区域,去除ROI中包含的非结节区域,能有效分割出基于灰度信息难以分割的毛玻璃型(GGO)结节;对于3种类型的结节区域:血管相连结节(VPN)、毛玻璃型结节(GG0)、孤立性结节(SPN)分割平均准确率分别为92.80%、86.13%、95.08%。     

基于区域划分的多特征纹理图像分割

由于纹理图像的复杂性和多样性,仅依靠传统的单一特征实现纹理图像分割无法满足其对分割精度的要求。本文提出结合区域划分的多特征纹理图像分割方法。首先,依据像素灰度的空间相关性定义多个纹理特征;然后利用区域划分将图像域划分成不同子区域,待分割同质区域由这些子区域拟合而成;通过分别定义多个特征图像的同质区域之间的异质性势能函数和刻画各子区域邻域关系势能函数来定义全局势能函数,并构建非约束吉布斯概率分布,从而建立纹理分割模型;最后,采用M-H算法采样上述概率分布,从而获得最优图像分割结果。分别对模拟纹理图像、遥感图像、自然纹理图像和SAR海冰图像进行了分割实验,并与利用单一特征得到的分割结果进行对比分析,定性和定量的测试结果验证了算法的有效性。     

Automated techniques for blood vessels segmentation through fundus retinal images: A review

Retina is the interior part of human's eye, has a vital role in vision. The digital image captured by fundus camera is very useful to analyze the abnormalities in retina especially in retinal blood vessels. To get information of blood vessels through fundus retinal image, a precise and accurate vessels segmentation image is required. This segmented blood vessel image is most beneficial to detect retinal diseases. Many automated techniques are widely used for retinal vessels segmentation which is a primary element of computerized diagnostic systems for retinal diseases. The automatic vessels segmentation may lead to more challenging task in the presence of lesions and abnormalities. This paper briefly describes the various publicly available retinal image databases and various machine learning techniques. State of the art exhibited that researchers have proposed several vessel segmentation methods based on supervised and supervised techniques and evaluated their results mostly on publicly datasets such as digital retinal images for vessel extraction and structured analysis of the retina. A comprehensive review of existing supervised and unsupervised vessel segmentation techniques or algorithms is presented which describes the philosophy of each algorithm. This review will be useful for readers in their future research.     

Automated system for characterization and classification of malaria‐infected stages using light microscopic images of thin blood smears

In this paper, we propose a comprehensive image characterization cum classification framework for malaria‐infected stage detection using microscopic images of thin blood smears. The methodology mainly includes microscopic imaging of Leishman stained blood slides, noise reduction and illumination correction, erythrocyte segmentation, feature selection followed by machine classification. Amongst three‐image segmentation algorithms (namely, rule‐based, Chan–Vese‐based and marker‐controlled watershed methods), marker‐controlled watershed technique provides better boundary detection of erythrocytes specially in overlapping situations. Microscopic features at intensity, texture and morphology levels are extracted to discriminate infected and noninfected erythrocytes. In order to achieve subgroup of potential features, feature selection techniques, namely, F‐statistic and information gain criteria are considered here for ranking. Finally, five different classifiers, namely, Naive Bayes, multilayer perceptron neural network, logistic regression, classification and regression tree (CART), RBF neural network have been trained and tested by 888 erythrocytes (infected and noninfected) for each features’ subset. Performance evaluation of the proposed methodology shows that multilayer perceptron network provides higher accuracy for malaria‐infected erythrocytes recognition and infected stage classification. Results show that top 90 features ranked by F‐statistic (specificity: 98.64%, sensitivity: 100%, PPV: 99.73% and overall accuracy: 96.84%) and top 60 features ranked by information gain provides better results (specificity: 97.29%, sensitivity: 100%, PPV: 99.46% and overall accuracy: 96.73%) for malaria‐infected stage classification.     

Accurate automatic detection of acute lymphatic leukemia using a refined simple classification

An improved classification technique is presented to identify automatically the acute lymphatic leukemia (ALL) subtypes. An adaptive segmentation procedure is performed on peripheral blood smear images to extract the main features (10 geometric features) from the segmented images of white blood cell (WBC), nucleus, and cytoplasm. To show the importance of the different extracted features for the diagnostic accuracy, a comprehensive study is made on all the possible permutation cases of the features using powerful classifiers which are K‐nearest neighbor (KNN) at different metric functions, support vector machine (SVM) with different kernels, and artificial neural network (ANN). This procedure enables us to construct a feature map depending only on least number of features which lead to the highest diagnostic accuracy. It is found that the features map regarding the vacuoles in the cytoplasm and the regularity of the nucleus membrane gives the highest accurate results. The automatic classification for ALL subtypes based only on these two effective features is assessed using the receiver operating characteristic (ROC) curve and F ₁ ‐score measures. It is confirmed that the present technique is highly accurate, and saves the effort and time of training.     

双边特征聚合与注意力机制点云语义分割

机器视觉是环境感知的重要手段之一,是自动驾驶、机器人、工业检测等领域的研究热点,而点云数据的精细分析是其 中的一项关键技术。 针对大尺度真实场景点云数据分割精度低的问题,提出了一种适用于点云数据语义分割的网络结构。 首 先,构建了一个双边特征聚合结构,通过分别处理点云的几何信息和语义信息,达到充分利用点云特征信息的目的。 其次,使用 近邻特征的高维空间相关性计算点与点之间的相互作用,进行局部邻域的上下文信息增强。 提出了一种混合池化结构代替最 大值池化,减少信息损失,使用横向跨层池化连接来增强特征多样性。 最后,引入注意力机制提取全局特征,滤除尺度噪声,增 强特征在空间上的表现力。 实验结果表明,该方法在大尺度真实场景点云数据集 S3DIS 上的平均交并比为 68. 2% ,平均准确率 为 80. 7% ,比 PointNet 提高了 20. 6% 和 14. 5% ,客观指标优于已有的代表性方法。     

一种双分支结构的图像语义分割算法

图像语义分割需要精细的细节信息和丰富的语义信息,然而在特征提取阶段,连续下采样操作会导致图像中物体的空间细节信息丢失。为解决该问题,提出一种双分支结构语义分割算法,在特征提取阶段既能有效获取丰富的语义信息又能减少物体细节信息的丢失。该算法的一个分支使用浅层网络保留高分辨率细节信息有助于物体的边缘分割,另一个分支使用深层网络进行下采样获取语义信息有助于物体的类别识别,再将两种信息有效融合可以生成精确的像素预测。通过Cityscapes数据集和CamVid数据集上的实验验证,与现有语义分割算法相比,所提算法在较少的参数条件下,获得了较好的分割效果。