Home / Current Issue / Paper 1705719
Deep Learning for Eye Disease Detection with Confidence Estimation and Explainable AI
Subject area: Science,Engineering and Technology · Area of research: AI in Health
Abstract
This research made progress on not only identifying valid deep learning models that can detect various eye diseases but also making the diagnosis process easier for physicians. This study focuses on three eye diseases, Cataract, Glaucoma and Diabetic Retinopathy. These three diseases are the leading causes of eye disorders which have resulted in irreversible visual impairment. Also, these three diseases are at different locations of the eye, the detection of their symptoms and analysis of the fundus images have set up different challenges. Various models, including Convolutional Neural Network (CNN), transfer learning architectures such as VGG16, InceptionV3, ResNet152V2, InceptionResNetV2, and DenseNet201 were used for Eye disease detection from patients? retinal images. DenseNet201 excelled, achieving a notable accuracy of 86.14%, a precision of 90.52%, a recall of 82.58%, and an AUC of 97.95%. To enhance model robustness, Monte Carlo (MC) dropout was integrated to estimate uncertainty levels, revealing the model's favourable performance under uncertainty. Furthermore, Local Interpretable Model-agnostic Explanations (LIME) was employed to elucidate the model's decision-making process, providing insights into how predictions were derived. This comprehensive approach showcases the efficacy of combining diverse models, leveraging transfer learning, and employing uncertainty estimation and explainability techniques for accurate eye disease detection.
How to cite this paper
@article{1705719,
author = {Adeyinka Mayowa-Majaro},
title = {Deep Learning for Eye Disease Detection with Confidence Estimation and Explainable AI},
journal = {Iconic Research And Engineering Journals},
year = {2024},
volume = {7},
number = {10},
pages = {204-212},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1705719.pdf},
abstract = {This research made progress on not only identifying valid deep learning models that can detect various eye diseases but also making the diagnosis process easier for physicians. This study focuses on three eye diseases, Cataract, Glaucoma and Diabetic Retinopathy. These three diseases are the leading causes of eye disorders which have resulted in irreversible visual impairment. Also, these three diseases are at different locations of the eye, the detection of their symptoms and analysis of the fundus images have set up different challenges. Various models, including Convolutional Neural Network (CNN), transfer learning architectures such as VGG16, InceptionV3, ResNet152V2, InceptionResNetV2, and DenseNet201 were used for Eye disease detection from patients? retinal images. DenseNet201 excelled, achieving a notable accuracy of 86.14%, a precision of 90.52%, a recall of 82.58%, and an AUC of 97.95%. To enhance model robustness, Monte Carlo (MC) dropout was integrated to estimate uncertainty levels, revealing the model's favourable performance under uncertainty. Furthermore, Local Interpretable Model-agnostic Explanations (LIME) was employed to elucidate the model's decision-making process, providing insights into how predictions were derived. This comprehensive approach showcases the efficacy of combining diverse models, leveraging transfer learning, and employing uncertainty estimation and explainability techniques for accurate eye disease detection.},
month = {April},
}