Deep Learning for Diabetic Retinopathy In Fundus Images

EasyChair Preprint 9330

Abstract

Clinically, using fundus pictures for predicting and detecting blind illnesses such as diabetic retinopathy (DR) is crucial. Deep learning (DL) is becoming a more common and promising technique in the different applications of DR, such as prediction, detection, classification, and disease diagnosis. Developing a review paper to analyze the DL techniques and their performance in the field is essential. We prepared a standard systematic review database including 341 publications. Accordingly, the main aim of the present review work is to present a systematic state-of-the-art by relying on PRISMA guidelines for the performance analysis of the DL in DR applications. The study has been shown in three main steps. The first step is to collect the database, the second step is to analyze the databases, and the last step is to conclude the study's main findings. According to the results, most studies employed accuracy as the most reliable and general evaluation metric for analyzing the DL techniques in different DR applications. Also, CNN has the most share of applications compared to other DL techniques. On the other hand, the best performance is related to the ensemble and advanced DL techniques. We'll also publish and regularly update the most recent discoveries in future studies to stay up with the quick technological improvements

Keyphrases: Diabetic Retinopathy, PRISMA, deep learning, machine learning, systematic review

@booklet{EasyChair:9330,
  author    = {Keyvan Rahimi and Rituraj Rituraj and Diana Ecker},
  title     = {Deep Learning for Diabetic Retinopathy In Fundus Images},
  howpublished = {EasyChair Preprint 9330},
  year      = {EasyChair, 2022}}