Deep learning: diabetic retinopathy detection using fundus image
Diabetic retinopathy (DR) is one of the leading causes of blindness worldwide. DR normally come with diabetes when it affects an individual. Since DR is one of the leading causes that cause blindness, early detection of it is much essential to prevent vision loss. However, the traditional method...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project / Dissertation / Thesis |
| Published: |
2025
|
| Subjects: | |
| Online Access: | http://eprints.utar.edu.my/6987/1/fyp_CS_2025_BJY.pdf http://eprints.utar.edu.my/6987/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Diabetic retinopathy (DR) is one of the leading causes of blindness worldwide. DR
normally come with diabetes when it affects an individual. Since DR is one of the
leading causes that cause blindness, early detection of it is much essential to prevent
vision loss. However, the traditional method of diagnosing DR is to rely on manual
examination of retinal images by ophthalmologists. It is a time-consuming and is
highly prone to error process since it is done by using manpower. Nowadays, with the
growing number of diabetic patients, there is an urgent need for an efficient and
automated solution for DR screening to prevent blindness due to DR, and therefore
this project is proposed. This project is aimed to develop a deep learning-based
system using Convolutional Neural Networks (CNNs) for the automated detection and
classification of DR. The dataset that is applied in this project is obtained from
Kaggle “Diabetic Retinopathy 224 x 224 Gaussian-Filtered” dataset, which consist
high-resolution fundus images across the five classes: No DR, Mild, Moderate, Severe
and Proliferate DR. Due to the significant class imbalance appear in the dataset, some
data augmentation techniques such as flipping, rotation, and zooming, along with
class weighting are applied to improve the performance. A DenseNet-201 model with
transfer learning from ImageNet was employed, enhanced with global average
pooling, batch normalization, dropout, and a softmax output layer. Some metrics were
used to evaluate the performance of the model such as accuracy, precision, recall, F1-
score and confusion matrix analysis. The experiment results show that the proposed
DenseNet-201 model achieves strong classification performance and shows promise
as a reliable and efficient tool for automated DR screening, supporting early detection
and reducing the workload of the ophthalmologists.
|
|---|