Home / Current Issue / Paper 1705807
Early Detection of Diabetic Retinopathy with Segmentation Model U-Net
Subject area: Science,Engineering and Technology · Area of research: Artificial Intelligence and Data Science
Abstract
Diabetic Retinopathy (DR) is a serious diabetes complication leading to irreversible vision loss, thus emphasizing the need for early detection and intervention. Leveraging advancements in deep learning, this study investigates the application of a segmentation model based on the U-Net architecture for the early detection of DR. The U-Net model, renowned for its efficacy in semantic segmentation tasks, offers a promising approach to accurately delineating retinal structures indicative of DR-related lesions. Through comprehensive experimentation and evaluation, including the comparison of a standard U-Net segmentation model and U-Net with a VGG16 pre-trained encoder, termed U-NetVGG16, the models' proficiency in achieving high accuracy, precision, recall, Intersection over Union (IoU), and F1-score metrics was demonstrated. U-NetVGG16 excelled, earning a notable IoU of 98.62%, an accuracy of 99.10%, a precision of 99.30%, a recall of 99.40%, and an f1-score of 99.30%. The results highlight the potential of deep learning-based segmentation models in revolutionizing diabetic eye care by facilitating automated and precise identification of DR-related abnormalities. This study contributes to advancing the field of early DR detection, aiming to mitigate the global burden of preventable vision impairment associated with this debilitating condition.
How to cite this paper
@article{1705807,
author = {Adeyinka Mayowa-Majaro},
title = {Early Detection of Diabetic Retinopathy with Segmentation Model U-Net},
journal = {Iconic Research And Engineering Journals},
year = {2024},
volume = {7},
number = {11},
pages = {337-343},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1705807.pdf},
abstract = {Diabetic Retinopathy (DR) is a serious diabetes complication leading to irreversible vision loss, thus emphasizing the need for early detection and intervention. Leveraging advancements in deep learning, this study investigates the application of a segmentation model based on the U-Net architecture for the early detection of DR. The U-Net model, renowned for its efficacy in semantic segmentation tasks, offers a promising approach to accurately delineating retinal structures indicative of DR-related lesions. Through comprehensive experimentation and evaluation, including the comparison of a standard U-Net segmentation model and U-Net with a VGG16 pre-trained encoder, termed U-NetVGG16, the models' proficiency in achieving high accuracy, precision, recall, Intersection over Union (IoU), and F1-score metrics was demonstrated. U-NetVGG16 excelled, earning a notable IoU of 98.62%, an accuracy of 99.10%, a precision of 99.30%, a recall of 99.40%, and an f1-score of 99.30%. The results highlight the potential of deep learning-based segmentation models in revolutionizing diabetic eye care by facilitating automated and precise identification of DR-related abnormalities. This study contributes to advancing the field of early DR detection, aiming to mitigate the global burden of preventable vision impairment associated with this debilitating condition.},
month = {May},
}