KLASIFIKASI PENYAKIT MATA PADA CITRA FUNDUS MENGGUNAKAN ALGORITMA CONVOLUTIONAL NEURAL NETWORK DENGAN ARSITEKTUR XCEPTION

The advancement of computer technology and artificial intelligence, particularly in image processing, has made significant contributions to the healthcare field, including eye disease detection using fundus images. Eye diseases such as diabetic retinopathy, glaucoma, and cataracts can lead to serious vision impairment if not diagnosed and treated in a timely manner. Fundus images, which are two-dimensional representations of the retina, are commonly used by doctors to detect abnormalities in the eyes. However, manual analysis methods relying on a doctor's experience can result in inter-observer errors. Therefore, a more reliable and consistent automated system is needed. This research aims to classify eye diseases in fundus images using a Convolutional Neural Network (CNN) algorithm with the Xception architecture. In this study, three testing scenarios were applied: (1) using Xception architecture as the base model, (2) data augmentation to enrich the training data variation, and (3) adding dropout and L2 regularization to reduce overfitting. The results from the third scenario showed a significant performance improvement, achieving a test accuracy of 72%, precision of 72.6%, recall of 72%, and an F1-Score of 71.8%. The implementation of data augmentation and regularization techniques proved effective in improving the model's generalization capability. This study demonstrates that CNN with Xception architecture can be a potential solution for automatic eye disease detection from fundus images.