Classification of Retinal Pathology via OCT Images using Convolutional Neural Network
Classification of Retinal Pathology via OCT Images using Convolutional Neural Network
复制标题
使用卷积神经网络通过 OCT 图像对视网膜病理进行分类
DOI：
10.1109/cosite52651.2021.9649630
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
S. Rizal
中科院分区：
文献类型：
--
作者：
Dewi Annisa Anam;L. Novamizanti;S. Rizal
光学相干断层扫描 (OCT) 是一种医学成像技术，用于检测黄斑中发生的病理情况。手动分析过程在时间和诊断准确性方面往往效率较低。本研究提出了一种基于 OCT 视网膜图像的广义黄斑视网膜病理学自动分类系统，使用具有 EfficientNet 架构的卷积神经网络 (CNN)。在预处理阶段，对图像进行了三种类型的信号处理分析，即高斯滤波器、对比度有限自适应直方图均衡（CLAHE）和Gabor滤波器。本文还评估了两种不同的优化器，即自适应矩（Adam）和随机梯度下降（SGD）。 EfficientNet 模型使用组合缩放方法来平衡所有网络维度。实验结果表明，该模型的预处理CLAHE和Adam优化函数可以对四个标准视网膜黄斑病理类别进行分类。这四个类别包括年龄相关性黄斑变性（AMD）、脉络膜新生血管（CNV）和糖尿病性黄斑水肿（DME），准确度为90.60。 %，损失0.27。
Optical Coherence Tomography (OCT) is a medical imaging technique used to detect pathology that occurs in the macula. The manual analysis process tends to be less effective and efficient both in time and diagnostic accuracy. This study proposes an automatic classification system for generalized macular retinal pathology based on OCT retinal images using Convolutional Neural Network (CNN) with EfficientNet architecture. In the preprocessing stage, three types of signal processing are analyzed on the image, namely Gaussian Filter, Contrast Limited Adaptive Histogram Equalization (CLAHE), and Gabor Filter. This paper also evaluates two different optimizers, namely Adaptive Moment (Adam) and Stochastic Gradient Decent (SGD). The EfficientNet model works using a combined scaling method to balance all network dimensions. The experimental results show that the proposed model's preprocessing CLAHE and Adam's optimization function can classify four standard retinal macular pathology classes. The four classes include Age-Related Macular Degeneration (AMD), Choroidal Neovascularization (CNV), and Diabetic Macular Edema (DME), with an accuracy of 90.60. %, and a loss of 0.27.