Retinopathy Classification using Convolutional Neural Network Method with Adaptive Momentum Optimization and Applied Batch Normalization

Abstract

Retinopathy is a common eye disease in Indonesia, ranking fourth after cataracts, glaucoma, and refractive errors. It can be overcome by early diagnosis with optical coherence tomography (OCT), but this imaging technique takes much time. In this research, retinal imaging was carried out using an expert system. The expert system in this study was formed using the convolutional neural network (CNN or ConvNet) method. CNN is an algorithm of deep learning that uses convolution operations to process two-dimensional data, such as images and sounds. This research consisted of 4 stages: data collection, preprocessing, model design, and model testing. A CNN model was formed with three arrangements, consisting of two convolutional layers and one pooling layer. The ReLU activation function, zero padding, and batch normalization were used in all three formats. Then, the flattening process was carried out, and the Softmax activation function was used at the end of the architecture. The model was built using eight epochs, and optimization of Adaptive Momentum resulted in a 98.96% test data accuracy value. The result was considered good so that CNN could be used as an alternative in retinopathy diagnosis. Further research is suggested to use other optimizations or other model architectures.