<![CDATA[Diabetic retinopathy is a symptom caused by complications of diabetes that attack the eyes of sufferers. Spots on the sufferer’s retina are characteristic of the symptoms. The more spots, the more severe the diabetic retinopathy suffered. Researchers’ efforts to detect diabetic retinopathy with retinal images have begun to be developed with artificial intelligence technology, one of which is based on deep learning. The next difficulty is the poor quality of the retinal image, resulting in poor detection results. Therefore, this study proposes a comparative analysis of techniques to improve image processing accuracy for deep learning-based diabetic retinopathy detection. The data used is APTOS2019 data, which consists of 5 classes based on the severity of the disease. There are three techniques used: CLAHE, gamma correction, and Retinex. The deep learning architecture used is DenseNet121 and EfficientNetB0 because it has been widely used in medical image data. As a result, the combination of gamma correction and DenseNet121 produces the highest accuracy of 81.4%. While the lowest accuracy is obtained from the combination using Retinex. The best overall architecture is EfficientNetB0, with an average accuracy of 81.9%. Furthermore, this study can be used to improve diabetic retinopathy images so that detection can be done as early as possible.]]>
Copyrights © 2025
