Page 167 - Kaleidoscope Academic Conference Proceedings 2024

AI-DRIVEN EARLY PREDICTION OF EYE DISORDER

1
Shailendra, Sagar and Megha, Agarwal 2
1
Department of Telecommunication
2
Jaypee Institute of Information Technology

ABSTRACT and early diagnosis. Data shows that glaucoma often
progresses without symptoms until significant vision loss
Diabetic retinopathy (DR) arises when diabetes mellitus occurs. Early detection through regular eye exams is crucial
damages the blood vessels in the retina, the crucial to prevent irreversible damage. The most common form of
part of the eye responsible for image capture. Without glaucoma, primary open-angle glaucoma (POAG), is often
timely intervention, this damage can lead to vision associated with elevated intraocular pressure. This pressure
impairment or even blindness. Given the cost and damages the optic nerve over time, leading to vision loss.
time constraints associated with manual examination of Predicting eye disorders is crucial for several reasons, ranging
fundus images by trained medical professionals, there’s from preventing vision loss to reducing healthcare costs.
a pressing need for automated DR detection systems. Predictive analytics allow for early detection and timely
AI-enabled teleophthalmology leveraging 5G/6G technology intervention, which can prevent or slow down the progression
can enhance eye care accessibility, particularly in remote of these diseases. Early treatment of eye conditions generally
regions. Transfer learning plays a pivotal role in medical results in better outcomes. For example, early-stage DR can
image analysis by facilitating the development of accurate often be managed with better blood sugar control and laser
and efficient models, especially when data and computational treatments, significantly reducing the risk of severe vision
resources are limited. This paper presents a modification of impairment. Treating eye disorders in their early stages is
the VGG16 network for DR detection, incorporating a feature often less expensive than managing advanced disease. Early
selection technique. By fine-tuning network parameters intervention can reduce the need for more invasive and costly
and selecting superior features, our proposed approach treatments or surgeries.
significantly enhances performance. Classification using AI-driven personalized e-services can make significant
machine learning models yields promising results over a advancements in predictive analytics and early diagnosis
benchmark DR dataset, with our modified VGG16 network for eye disorders. By leveraging machine learning, image
achieving an impressive accuracy of 98.4%, surpassing analysis, and big data, these services can provide valuable
the standard VGG16 and ResNet18 network along with tools for detecting and managing eye conditions. AI
state-of-the-art methods. algorithms can detect DR from retinal images with accuracy
comparable to that of experienced ophthalmologists. AI
Keywords - Artificial Intelligence, Disease Prediction, models can also detect glaucomatous damage by analyzing
Retinal Disease, VGG16 optic nerve images and visual field tests, even before
significant symptoms appear.
1. INTRODUCTION AI algorithms analyze thousands of images quickly and
accurately, reducing the burden on human specialists and
According to the World Health Organization (WHO), eye
decreasing the time to diagnosis. It provides consistent results
disorders such as diabetic retinopathy (DR) and glaucoma are
that are not influenced by human factors such as fatigue or
major public health concerns due to their potential to cause
subjective interpretation, leading to more reliable diagnostics.
significant visual impairment and blindness. Approximately
The objective of this paper is to evaluate VGG16 network
463 million people globally have diabetes, and about
and modify it by incorporating feature selection mechanism
one-third of these individuals have some form of DR.
and automatically diagnose DR from fundus images more
This translates to around 150 million people worldwide
precisely.
affected by DR. Roughly one-third of those with DR
have vision-threatening forms of the condition, amounting
to around 50 million people at risk of significant visual 2. LITERATURE SURVEY
impairment. Data shows that the longer someone has
diabetes, the higher their risk of developing DR. Nearly all The study in [1] reviews 94 articles from 2018-2023
patients with diabetes for over 20 years will experience some highlighting the effectiveness of transfer learning and popular
degree of retinopathy. CNN models (ResNet, VGGNet) in DR classification using
Over 60 million people are affected by glaucoma worldwide, datasets like APTOS 2019 and EyePACS. They concluded
with around 4.5 million individuals blind due to the condition. that despite advancements in deep learning, challenges
These figures underscore the importance of regular screening remain in integrating these systems into clinical practice,

978-92-61-39091-4/CFP2268P @ITU 2024 – 123 – Kaleidoscope