Annu Int Conf IEEE Eng Med Biol Soc. 2022 Jul;2022:3858-3861. doi: 10.1109/EMBC48229.2022.9871322.
Optical Coherence Tomography (OCT) is a non-invasive imaging technology that is widely applied for the diagnosis of retinal pathologies. In general, the structural information of retinal layers plays an important role in the diagnosis of various eye diseases by ophthalmologists. In this paper, by focusing on this information, we first introduce a new layer called the edge convolutional layer (ECL) to accurately extract the retinal boundaries in different sizes and angles with a much smaller number of parameters than the conventional convolutional layer. Then, using this layer, we propose the ECL-guided convolutional neural network (ECL-CNN) method for the automatic classification of the OCT images. For the assessment of the proposed method, we utilize a publicly available data comprising 45 OCT volumes with 15 age-related macular degeneration (AMD), 15 diabetic macular edema (DME), and 15 normal volumes, captured by using the Heidelberg OCT imaging device. Experimental results demonstrate that the suggested ECL-CNN approach has an outstanding performance in OCT image classification, which achieves an average precision of 99.43% as a three-class classification work. Clinical Relevance - The objective of this research is to introduce a new approach based on CNN for the automated classification of retinal OCT images. Clinically, the ophthalmologists should manually check each cross-sectional B-scan and classify retinal pathologies from B-scan images. This manual process is tedious and time-consuming in general. Hence, an automatic computer-assisted technique for retinal OCT image classification is demanded.
光学相干断层扫描（OCT）是一种广泛应用于视网膜病变诊断的非侵入性成像技术。一般来说，眼科医生通过关注视网膜各层的结构信息，对各种眼病进行诊断。在本文中，我们首先介绍了一种新的层，称为边缘卷积层（ECL），它可以用比传统卷积层小得多的参数，在不同大小和角度下准确地提取视网膜边界。然后，我们使用这个层提出了 ECL 引导的卷积神经网络（ECL-CNN）方法，用于 OCT 图像的自动分类。为了评估所提出的方法，我们利用了一个公开的数据集，其中包含 45 个 OCT 体积，分别为 15 个年龄相关性黄斑变性（AMD）、15 个糖尿病性黄斑水肿（DME）和 15 个正常体积，这些数据是使用海德堡 OCT 成像设备采集的。实验结果表明，所提出的 ECL-CNN 方法在 OCT 图像分类中具有出色的性能，作为一个三分类工作，其平均精度达到了 99.43%。临床相关性 - 本研究的目的是引入一种基于 CNN 的新方法，用于自动分类视网膜 OCT 图像。在临床上，眼科医生需要手动检查每个横断面 B 扫描，并从 B 扫描图像中分类视网膜病变。这个手动过程通常既繁琐又耗时。因此，需要一种用于视网膜 OCT 图像分类的自动计算机辅助技术。
