International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 05 | May 2024
p-ISSN: 2395-0072
www.irjet.net
HYPERPARAMETER-TUNED DENSENET121 FOR DIABETIC RETINOPATHY CLASSIFICATION: A DEEP LEARNING APPROACH P. Saranya 1, Dr. C. Vennila2 1 Assistant Professor, Dept. of Computer Science & Engg., Government College of Engg. Srirangam, Tamilnadu,
India
2 Professor, Dept of Electronics & Communication Engg., Saranathan college of Engineering, Tamilnadu, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In this study, we employ a hyperparameter-tuned
these new vessels continue to grow until they rupture, leading to bleeding in the vitreous cavity or tearing the retina, ultimately resulting in vision loss due to tissue expansion. The second issue is plasma leakage, which involves lipid exudation that deposits fat in the macula, altering its structure and leading to vision impairment.
DenseNet121 architecture within a Convolutional Neural Network (CNN) framework to analyze fundus oculi images for predicting the presence and severity of Diabetic Retinopathy (DR). Diabetes, a condition characterized by elevated blood sugar levels, can lead to DR, a significant cause of vision impairment and blindness, especially among older individuals. Early detection of DR is crucial for timely intervention and treatment. Our model is trained and evaluated using a dataset comprising labeled fundus oculi images, each annotated with the severity of DR. Leveraging hyperparameter tuning, specifically optimizing the learning rate and dropout rate, we enhance the performance of the DenseNet121-based CNN model. Through rigorous experimentation, we demonstrate the effectiveness of our approach in accurately classifying DR severity levels, thus contributing to early diagnosis and management strategies for this sight-threatening condition.
After examining the retina's fundus, diabetic retinopathy can be classified from its mildest to most severe stages. The two primary forms of the condition are NonProliferative Diabetic Retinopathy (NPDR) and Proliferative Diabetic Retinopathy (PDR). NPDR is further divided into three subcategories: mild, moderate, and severe, as illustrated in Figure 1.
Key Words:
Diabetic Retinopathy, Deep Learning, Convolutional Neural Network, Transfer Learning.
1.INTRODUCTION Diabetic Retinopathy results from damage to the blood vessels in the retina caused by diabetes. Individuals with diabetes often experience some degree of retinal damage. The affected blood vessels can swell, leak, or promote the growth of new blood vessels. The loss of pericytes, which are contractile cells that envelop capillary endothelial cells in the body's venules, contributes to capillary damage. This damage occurs due to high levels of glucose in the blood, which clump together in the capillaries and impede blood flow, a condition known as ischemia. Microaneurysms, resulting from the diminished blood flow caused by the deterioration of these blood vessels, are saccular enlargements at the venous end of retinal capillaries. This process compromises the arteries' impermeability, leading to leaks such as bleeding or lipid exudation.
Fig-1: Stages of Diabetic Retinopathy Deep learning (DL) is a well-established technique that automatically extracts features from images through a convolutional neural network's layer stack. These features enable the classification of image contents by identifying specific patterns. In this study, we propose a DL model to categorize retina fundus images and identify diabetic retinopathy (DR) across all stages. We utilize DenseNet121, a type of Convolutional Neural Network, to distinguish between healthy eyes and those affected by proliferative diabetic retinopathy.
2. RELATED WORKS Numerous studies have explored the detection and classification of diabetic retinopathy (DR) using various methods. One of the pioneering works in this field is by Cree et al. [2], who developed a system that used handengineered features and empirically determined parameters to classify digitized retina fundus images.
Ischemia in the retina leads to two major complications. The first issue involves the synthesis of the cytokine protein VEGF, which promotes the formation of new blood vessels (neovascularization) from existing ones. This protein can cause problems by proliferating on the surface of the vitreous humor and retina. Due to insufficient blood flow,
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 2140