Neural Network Analysis for the detection of glaucomatous damage

Glaucoma is a major cause of blindness and is prevalent among Asian populations. Therefore, early detection is of paramount importance in order to let patients have early treatments. One prominent indicator of glaucomatous damage is the Retinal Nerve Fiber Layer (RNFL) profile. In this paper, the performance of artificial neural network models in identifying RNFL profile of glaucoma suspect and glaucoma subjects is studied. RNFL thickness was measured using optical coherence tomography (Stratus OCT). Inputs to the neural network consisted of regional RNFL thickness measurements over 12 clock hours. Sensitivity and specificity for glaucoma detection will be compared by the area under the Receiver Operating Characteristic Curve (AROC). The results show that artificial neural network coupled with the OCT technology enhances the diagnostic accuracy of optical coherence tomography in differentiating glaucoma suspect and glaucoma from normal individuals.     

Automated segmentation of macular layers in OCT images and quantitative evaluation of performances

Optical coherence tomography (OCT) allows high-resolution and noninvasive imaging of the structure of the retina in humans. This technique revolutionized the diagnosis of retinal diseases in routine clinical practice. Nevertheless, quantitative analysis of OCT scans is yet limited to retinal thickness measurements. We propose a novel automated method for the segmentation of eight retinal layers in these images. Our approach is based on global segmentation algorithms, such as active contours and Markov random fields. Moreover, a Kalman filter is designed in order to model the approximate parallelism between the photoreceptor segments and detect them. The performance of the algorithm was tested on a set of retinal images acquired in-vivo from healthy subjects. Results have been compared with manual segmentations performed by five different experts, and intra and inter-physician variability has been evaluated as well. These comparisons have been carried out directly via the computation of the root mean squared error between the segmented interfaces, region-oriented analysis, and retrospectively on the thickness measures derived from the segmentations. This study was performed on a large database including more than seven hundred images acquired from more than one hundred healthy subjects.     

A marker-based watershed method for X-ray image segmentation

Digital X-ray images are the most frequent modality for both screening and diagnosis in hospitals. To facilitate subsequent analysis such as quantification and computer aided diagnosis (CAD), it is desirable to exclude image background. A marker-based watershed segmentation method was proposed to segment background of X-ray images. The method consisted of six modules: image preprocessing, gradient computation, marker extraction, watershed segmentation from markers, region merging and background extraction. One hundred clinical direct radiograph X-ray images were used to validate the method. Manual thresholding and multiscale gradient based watershed method were implemented for comparison. The proposed method yielded a dice coefficient of 0.964 ± 0.069, which was better than that of the manual thresholding (0.937 ± 0.119) and that of multiscale gradient based watershed method (0.942 ± 0.098). Special means were adopted to decrease the computational cost, including getting rid of few pixels with highest grayscale via percentile, calculation of gradient magnitude through simple operations, decreasing the number of markers by appropriate thresholding, and merging regions based on simple grayscale statistics. As a result, the processing time was at most 6 s even for a 3072 × 3072 image on a Pentium 4 PC with 2.4 GHz CPU (4 cores) and 2G RAM, which was more than one time faster than that of the multiscale gradient based watershed method. The proposed method could be a potential tool for diagnosis and quantification of X-ray images.     

Computer Vision Techniques Applied for Diagnostic Analysis of Retinal OCT Images: A Review

The retina is a tiny layer at the posterior pole of an eye and is made up of tissues sensitive to light, these tissues generate nerve signals that pass through the optic nerve to the brain. A retinal disorder occurs when the retina malfunctions; glaucoma, diabetic retinopathy and pathologic myopia are retinal disorders and principal causes of blindness worldwide. These retinal disorders are often diagnosed and treated by an ophthalmologist. However, to accurately assess a retinal disease, ophthalmologist would need qualitative and quantitative analysis of the disease, it’s early and current statistics, but acquisition of these measurements are not possible through manual techniques, there should be automated computer aided diagnosis (CAD) systems to assist ophthalmologists. In this comprehensive review, an analysis and evaluation has been performed of different computer vision and image processing approaches applied to OCT images for automatic diagnosis of retinal disorders. We also reported disease causes, symptoms and pathologies manifestations within OCT images, which can serve as baseline knowledge for development of an automated CAD system. Hence, this disease specific review offers a good understanding to analyze visual impairments from retinal OCT images which will help researcher to design enhanced therapeutic systems for retinal disorders.     

结合PCANet与线性判别分析的视网膜光学相干断层扫描图像分类

目的 主成分分析网络（PCANet）能提取图像的纹理特征,线性判别分析（LDA）提取的特征有类别区分性。本文结合这两种方法的优点,提出一种带线性判别分析的主成分分析网络（PCANet-LDA）,用于视网膜光学相干断层扫描（OCT）图像中的老年性黄斑变性（AMD）、糖尿病性黄斑水肿（DME）及正常（NOR）这3类的全自动分类。方法 PCANet-LDA算法是在PCANet的基础上添加了LDA监督层,该层加入了类标签对特征进行监督投影。首先,对OCT视网膜图像进行去噪、二值化及对齐裁剪等一系列预处理,获得感兴趣的视网膜区域;然后,将预处理图像送入一个两层的PCA卷积层,训练PCA滤波器组并提取图像的PCA特征;接着,将PCA特征送入一个非线性输出层,通过二值散列和块直方图等处理,得到图像的特征;之后,将带有类标签的图像特征送入一个LDA监督层,学习LDA矩阵并用其对图像特征进行投影,使特征具有类别区分性;最后,将投影的特征送入线性支持向量机（SVM）中对分类器进行训练和分类。结果 实验分别在医院临床数据集和杜克数据集上进行,先对OCT图像预处理进行前后对比实验,然后对PCANet特征提取的有效性进行分析,最后对PCANet算法、ScSPM算法以及提出的PCANet-LDA3种分类算法的分类效果进行对比实验。在临床数据集上,PCANet-LDA算法的总体分类正确率为97.20%,高出PCANet算法3.77%,且略优于ScSPM算法;在杜克数据集上,PCANet-LDA算法的总体分类正确率为99.52%,高出PCANet算法1.64%,略优于ScSPM算法。结论 PCANet-LDA算法的分类正确率明显高于PCANet,且优于目前用于2D视网膜OCT图像分类的先进的ScSPM算法。因此,提出的PCANet-LDA算法在视网膜OCT图像的分类上是有效且先进的,可作为视网膜OCT图像分类的基准算法。     

基于双注意力编码-解码器架构的视网膜血管分割

眼底视网膜血管的分割提取对于糖尿病、视网膜病、青光眼等眼科疾病的诊断具有重要的意义。针对视网膜血管图像中的血管难以提取、数据量较少等问题,文中提出了一种结合注意力模块和编码-解码器结构的视网膜血管分割方法。首先对编码-解码器卷积神经网络的每个卷积层添加空间和通道注意力模块,加强模型对图像特征的空间信息和通道信息(如血管的大小、形态和连通性等特点)的利用,从而改善视网膜血管的分割效果。其中,空间注意力模块关注于血管的拓扑结构特性,而通道注意力模块关注于血管像素点的正确分类。此外,在训练过程中采用Dice损失函数解决了视网膜血管图像正负样本不均衡的问题。在3个公开的眼底图像数据库DRIVE,STARE和CHASE_DB1上进行了实验,实验数据表明,所提算法的准确率、灵敏度、特异性和AUC值均优于已有的视网膜血管分割方法,其AUC值分别为0.9889,0.9812和0.9831。实验证明,所提算法能够有效提取健康视网膜图像和病变视网膜图像中的血管网络,能够较好地分割细小血管。     

Generation of high-resolution East Antarctic landfast sea-ice maps from cloud-free MODIS satellite composite imagery

A method to generate high spatio-temporal resolution maps of landfast sea ice from cloud-free MODIS composite imagery is presented. Visible (summertime) and thermal infrared (wintertime) cloud-free 20-day MODIS composite images are used as the basis for these maps, augmented by AMSR-E ASI sea-ice concentration composite images (when MODIS composite image quality is insufficient). The success of this technique is dependent upon efficient cloud removal during the compositing process. Example wintertime maximum (~ 374,000 km²) and summertime minimum (~ 112,000 km²) fast-ice maps for the entire East Antarctic coast are presented. The summertime minimum map provides the first high-resolution indication of multi-year fast-ice extent, which may be used to help assess changes in Antarctic sea-ice volume. The 2σ errors in fast-ice extent are estimated to be ± 2.98% when ≥ 90% of the fast-ice pixels in a 20-day period are classified using the MODIS composite, or ± 8.76 otherwise (when augmenting AMSR-E or the previous/next MODIS composite image is used to classify > 10% of the fast ice). Imperfect composite image quality, caused by persistent cloud, inaccurate cloud masking or a highly dynamic fast-ice edge, was the biggest impediment to automating the fast-ice detection procedure.     

基于EM算法的眼底OCT图像反卷积去模糊技术

光学相干层析成像技术（Optical coherence tomography,OCT）在视网膜检查中十分重要,然而在获取OCT图像时眼球运动或者散焦作用都可能引起图像的模糊,从而为临床诊断造成困难。因此,从模糊OCT图像中恢复出清晰图像的去模糊技术研究至关重要。本文结合OCT成像原理,提出了一种基于最大期望（Expectation-maximization,EM）算法的OCT图像反卷积技术。该技术能够在一定程度上抑制OCT模糊图像中异常值对复原图像的干扰,从而有效去除OCT图像中的模糊。将本文技术与多种现有广义图像去模糊技术进行了实验比较,结果表明本文提出的复原OCT图像的反卷积算法在眼底OCT图像去模糊的细节恢复方面效果较好。     

Accuracy and uncertainty of thermal-infrared remote sensing of stream temperatures at multiple spatial scales

Stream temperature is an important indicator of water quality, particularly in regions where endangered fish populations are sensitive to elevated water temperature. Regional assessment of stream temperatures from the ground is limited by sparse sampling in both space and time. Remotely sensed thermal-infrared (TIR) images are able to make spatially distributed measurements of the radiant skin temperature of streams. We quantify and discuss the accuracy and uncertainty limits to recovering stream temperatures in the Pacific Northwest for a range of stream widths (10-500 m), and TIR pixel sizes (5-1000 m) from remotely sensed airborne and satellite TIR images. Among locations with more than three pixels across the stream, the image temperature overestimated the in-stream temperature on average by 1.2 °C, which is 7% of the in-stream temperature (standard error (SE) of 0.2 °C, n = 21). The corresponding uncertainty (band weighted standard deviation in image temperature) for these locations averaged ± 0.3 °C (SE < 0.1 °C, n = 21) which is 2% of in-stream temperatures. This overestimation by the image temperatures is likely to be due to thermal stratification between the stream surface and the location of the in-stream temperature measurements deeper in the water column. For streams with one to three pixels across, mixing with bank elements increased the overestimation by image temperatures to 2.2 °C (SE = 0.3 °C, n = 23) on average (13% of in-stream temperatures), and the uncertainty increased to ± 0.4 °C (SE = 0.1 °C, n = 23) which is 2% of in-stream temperatures. For a fraction of a pixel across the stream the overestimation by image temperatures was 7.6 °C (SE = 1.2 °C, n = 23) on average (45% of in-stream temperatures), and the uncertainty was ± 0.5 °C (SE = 0.1 °C, n = 23) which is 3% of in-stream temperatures. These results show that reliable satellite TIR measurement of stream temperatures is limited to large rivers (∼180-m across for Landsat ETM+), unless novel unmixing algorithms are used effectively.     

双层水平集描述眼底图像视杯视盘分割

目的 青光眼是一种可导致视力严重减弱甚至失明的高发眼部疾病。在眼底图像中,视杯和视盘的检测是青光眼临床诊断的重要步骤之一。然而,眼底图像普遍是灰度不均匀的,眼底结构复杂,不同结构之间的灰度重叠较多,受到血管和病变的干扰较为严重。这些都给视盘与视杯的分割带来很大挑战。因此,为了更准确地提取眼底图像中的视杯和视盘区域,提出一种基于双层水平集描述的眼底图像视杯视盘分割方法。方法 通过水平集函数的不同层级分别表示视杯轮廓和视盘轮廓,依据视杯与视盘间的位置关系建立距离约束,应用图像的局部信息驱动活动轮廓演化,克服图像的灰度不均匀性。根据视杯与视盘的几何形状特征,引入视杯与视盘形状的先验信息约束活动轮廓的演化,从而实现视杯与视盘的准确分割。结果 本文使用印度Aravind眼科医院提供的具有视杯和视盘真实轮廓注释的CDRISHTI-GS1数据集对本文方法进行实验验证。该数据集主要用来验证视杯及视盘分割方法的鲁棒性和有效性。本文方法在数据集上对视杯和视盘区域进行分割,取得了67.52%的视杯平均重叠率,81.04%的视盘平均重叠率,0.719的视杯F1分数和0.845的视盘F1分数,结果优于基于COSFIRE（combination of shifted filter responses）滤波模型的视杯视盘分割方法、基于先验形状约束的多相Chan-Vese（C-V）模型和基于聚类融合的水平集方法。结论 实验结果表明,本文方法能够有效克服眼底图像灰度不均匀、血管及病变区域的干扰等影响,更为准确地提取视杯与视盘区域。