Enhancement of Diabetic Retinopathy Prediction Based Capsule Network and DenseNet-121 Utilizing Wavelet Decomposition

Abstract

Diabetic retinopathy is a health condition associated with the leaking of blood vessels in the retina, potentially resulting in vision impairment. This medical condition impacts a broad population of people, especially individuals with diabetes. The primary challenge with this medical condition is its asymptomatic early phase, which makes its detection at an early stage a very challenging task, especially since its treatment is based on the early detection task. In this study, we will provide a novel automatic method based on DenseNet-121 and capsule network designs for the severity levels and the early detection of diabetic retinopathy, in addition to the discrete wavelet transform in the preprocessing step. However, the proposed approach generated an advanced image using the discrete wavelet transformation technique to feed it to the model. The model starts with the DenseNet-121 architecture and transfers the pieces of information to the capsule network to make decisions. We used the APTOS dataset to validate our approach, and the outcome was a good accuracy rating of 86.72%.