Balakrishnan Umarani, Venkatachalapathy Krishnamurthi, Marimuthu Girirajkumar S
Department of Electronics and Communication Engineering, Trichy Engineering College, Tiruchirappalli- 621 132, Tamilnadu, India.
Curr Diabetes Rev. 2015;11(3):182-90. doi: 10.2174/1573399811666150330150038.
Diabetic Retinopathy (DR) is an eye disease, which may cause blindness by the upsurge of insulin in blood. The major cause of visual loss in diabetic patient is macular edema. To diagnose and follow up Diabetic Macular Edema (DME), a powerful Optical Coherence Tomography (OCT) technique is used for the clinical assessment. Many existing methods found out the DME affected patients by estimating the fovea thickness. These methods have the issues of lower accuracy and higher time complexity. In order to overwhelm the above limitations, a hybrid approaches based DR detection is introduced in the proposed work. At first, the input image is preprocessed using green channel extraction and median filter. Subsequently, the features are extracted by gradient-based features like Histogram of Oriented Gradient (HOG) with Complete Local Binary Pattern (CLBP). The texture features are concentrated with various rotations to calculate the edges. We present a hybrid feature selection that combines the Particle Swarm Optimization (PSO) and Differential Evolution Feature Selection (DEFS) for minimizing the time complexity. A binary Support Vector Machine (SVM) classifier categorizes the 13 normal and 75 abnormal images from 60 patients. Finally, the patients affected by DR are further classified by Multi-Layer Perceptron (MLP). The experimental results exhibit better performance of accuracy, sensitivity, and specificity than the existing methods.
糖尿病性视网膜病变（DR）是一种眼部疾病，可能由于血液中胰岛素水平升高而导致失明。糖尿病患者视力丧失的主要原因是黄斑水肿。为了诊断和随访糖尿病性黄斑水肿（DME），一种强大的光学相干断层扫描（OCT）技术被用于临床评估。许多现有方法通过估计中央凹厚度来找出受DME影响的患者。这些方法存在准确性较低和时间复杂度较高的问题。为了克服上述局限性，在所提出的工作中引入了一种基于混合方法的DR检测。首先，使用绿色通道提取和中值滤波器对输入图像进行预处理。随后，通过基于梯度的特征（如具有完整局部二值模式（CLBP）的方向梯度直方图（HOG））来提取特征。纹理特征通过各种旋转进行集中以计算边缘。我们提出了一种结合粒子群优化（PSO）和差分进化特征选择（DEFS）的混合特征选择方法，以最小化时间复杂度。一个二元支持向量机（SVM）分类器对来自60名患者的13张正常图像和75张异常图像进行分类。最后，受DR影响的患者通过多层感知器（MLP）进一步分类。实验结果表明，与现有方法相比，该方法在准确性、敏感性和特异性方面具有更好的性能。
