A
REVIEW OF DEEP LEARNING APPLICATIONS IN OPHTHALMOLOGY
Dr. Pawan Thakur
Asst. Professor,
Department of B.Tech(CSE)
School of Computer
Science Engineering and Technology,
Government College
Dharamshala (H.P.), India.
Ishita*,
Sonali, Sejal
School of Computer
Science Engineering and Technology,
Government College
Dharamshala (H.P.), India.
(Corresponding
author: Ishita * ishita062003@gmail.com,
9816293025)
ABSTRACT
Early detection of eye disease is critical to saving
vision and preventing additional injury. Many types of eye diseases like
diabetic retinopathy and glaucoma can be asymptotic or can have few symptoms in
their early development. By the time symptoms have developed, the diseases may
have advanced to a point where it’s much more difficult to treat. Timely
discovery also allows for early commencement of suitable treatment that could
assist in controlling the health issue and preventing degeneration in vision.
This is the reason early detection of these ailments becomes so important
because individuals can lose their sight. Machine learning (ML), deep learning
(DL), and advanced imaging methods have greatly improved detection rates in eye
disease. This project uses CNN (Convolutional Neural Network) to classify the
eye type of people using deep learning.
Keywords: CNN,
Disease Detection, deep learning.
I.
INTRODUCTION
A subset of machine learning and AI
wherein artificial neural networks are used to interpret, process, and analyze
large datasets in a way reminiscent of human cognition. Deep learning, in
particular, uses many layers of interconnected nodes, referred to as deep
neural networks, to allow systems to autonomously learn complex patterns and
make intelligent decisions. Deep learning (DL) has transformative technology in
the field of medical imaging of eye disease detection and diagnosis. Common eye
disorders such as Diabetic retinopathy (DR), glaucoma, cataracts, and
age-related macular degeneration (AMD) often progress silently, which makes
early detection imperative to preventing loss of vision or blindness. The
classical diagnostic method depends on expert assessment, which often can be
lengthy and variable. DL-based methods provide automated and accurate solutions
by capturing the underlying complex patterns in the images. Deep learning by
detecting eye disease physically lowers the chances of severe vision loss and
blindness and enables the necessary treatment in time which safeguards the
vision and enhances the quality the life of the patients. Automation increases
accessibility, especially in rural areas, through consistent and efficient
screens that do not depend on specialists alone.
II.
RELATED WORK
A prospective review of deep
learning techniques employed in OCT imaging data for the identification of
intraocular lens and refractive surgical procedures (Ginsburg, 2006) used
machine-learning algorithms to pinpoint specific eye diseases. After that, different
machine learning approaches were used for detecting specific eye diseases like
age-related macular degeneration) and automatic diabetic retinopathy detection
and automated optical disc localization through image classification (Farooq
and Sattar, 2015). The authors (Wei Ting et al., 2017) design a deep learning
system diagnostic for detecting referrable AMD, their CNN lacks a step for
image processing that includes filters to extract deep features of OCT images.
This research (Burlina et al., 2017) reported high diagnostic accuracy by
pre-segmenting the macula region and testing step. Used deep CNNs to detect two
types of AMD with an accuracy of 92% highlighting the effectiveness of CNNs in
specific diseases. The concept of segmentation was also applied to blood
vessels to allow the early detection of eye diseases. (Moccia et al., 2018)
used a machine learning method model that performed a vessel segmentation
process to extract features that lead to the prediction of eye diseases. The
authors (Grassmann et al., 2018) proposed a deep learning-based approach for
the classification of age-related macular degeneration (AMD) from color fundus
photography. This highlights the potential of deep learning models in
accurately identifying age-related eye diseases. In this paper authors (Prasad
et al., 2019) introduced a deep neural network model that can recognize
features associated with diabetic retinopathy and glaucoma in the early stages.
They aim to advise the patients to consult an ophthalmologist to be sure of the
presence of the mentioned eye disease. This paper (Chen et al., 2015) focused
on glaucoma detection utilizing deep convolutional neural networks (CNNs). They
employed strategies such as data augmentation and dropout to enhance
performance and mitigate overfitting. They utilized response-normalization
layers and overlapping-pooling layers for further improvement in disease
segmentation and classification. (Chen et al.,2015) presented an approach
specifically tailored for feature learning in glaucoma detection. Their method
utilized CNNs for features extraction, enabling the hierarchical representation
of fundus images. The work (Krishna et al., 2019) proposed a method for the
simultaneous detection of diabetic retinopathy (DR) and glaucoma using deep neural
networks. This system achieved an accuracy of 80% for automated disease
detection. The authors (Aun and Nazir et al.,2020) introduced a technique
specifically targeting diabetics-based eye diseases. They employed the fast
region-based convolutional neural networks (FRCNN)algorithm for disease
localization and utilized Fuzzy k-means (FKM) clustering for disease
segmentation post detection. The authors (Sarkari et al.,2020) conducted a
comprehensive survey on diabetic eye disease detection, covering various
aspects, including deep learning models and image processing techniques. This
work provides a valuable overview of the current state-of-the-art approaches in
the field, facilitating further research and development in diabetic eye
disease detection. The authors (Chelaramani et al., 2020) addressed multiple
tasks related to eye diseases using fundus images, including disease category
detection, subcategory detection, and textual diagnosis generation. Leveraging
ResNET models and multitask learning, they achieved promising accuracies of 86%
for category detection and 67% for subcategory detection. (Elkholy and Marzouk,
2024) applied CNNs to classify Optical Coherence Tomography (OCT) images into
four categories: Normal Retina, Diabetic Macular Edema (DME),Age-Related
Macular Degeneration(AMD). The model achieved an accuracy of 97% out performing
other models in the literature.
Table 1: Summary of Research Studies
on Eye Disease Detection Using Deep Learning Techniques
|
Title of
Paper |
Year of Publication |
Authors |
Deep
Learning Techniques/Model/Algo used |
Accuracy |
|
OCT imaging
data for detecting intraocular lenses and refractive surgery |
2006 |
Ginsburg |
OCT |
Not Mentioned |
|
Diabetic
retinopathy |
2015 |
Farooq and
Sattar |
Automation Optical disc localization |
Not Mentioned |
|
Deep learning
system diagnostic for detecting referable AMD |
2017 |
Wei Ting |
OCT , CNN,AMD |
Not Mentioned |
|
Diagnostic
accuracy by pre-segmenting the macula region
|
2017 |
Burlina |
CNN , AMD |
92% |
|
Machine Learning
methods model |
2018 |
Moccia |
Deep Learning
|
Not Mentioned
|
|
Deep Learning
based approaches for AMD |
2018 |
Grassmann |
Deep Learning
& AMD |
Not Mentioned
|
|
Deep Neural
Network Model |
2019 |
Prasad |
Deep Learning |
Not Mentioned |
|
Focused on
glaucoma Detection |
2015 |
Chen |
CNN |
Not Mentioned |
|
Detection of
diabetic retinopathy |
2019 |
Krishna |
Deep neural
networks |
Not Mentioned |
|
Diabetic
based eye disease |
2020 |
Aun, Nazir |
FRCNN |
Not Mentioned |
|
Deep learning
Model |
2020 |
Sarkari |
Deep learning
|
Not Mentioned |
|
Disease
Category Detection |
2020 |
Chelaramani |
ResNET |
86% category
detection, 67% sub-category detection |
|
Classify
Optical Coherence Tomography (OCT) |
2024 |
Elkholy and
Marzouk |
AMD , OCT |
97% |
III. RESEARCH GAP
This
literature review presents compressive studies in eye disease detection using
deep learning models. There are many research gaps in these papers, including:
1. Few
research studies work with the restricted set of eye conditions, including
diabetic retinopathy, glaucoma, and age-related macular degeneration, despite
the existence of multiple other possible diseases.
2. The
success of deep learning models typically depends on matching particular
datasets and struggles to function equally well across different setups.
3. Deep
learning models face current criticism because their unpredictable
decision-making process makes prediction interpretation challenging for
clinicians to trust.
4. The
models designed for early detection tend to deliver poor results when achieving
high accuracy for multiple disease types simultaneously.
5. A lack
of exists on how deep learning models can smoothly integrate into the
operational framework of healthcare.
IV. IMPLICATION
The results from this study establish multiple consequences
for detecting eye diseases:
1. Debug Learning algorithms demonstrate high specificity
levels in identifying three primary vision-threatening eye conditions,
including diabetic retinopathy (DR), glaucoma, macular degeneration.
2. Automated systems with DL eliminate time consumption and
resource usage and minimize human mistakes that frequently lead to errors and
labor-intensive processes in manual diagnosis procedures.
3. Early detection becomes possible through DL because it
analyzes the minor image patterns in the eye that human expert might overlook.
4. Object-centered monitoring methods built into DL systems
help analyze disease evolution or reverse rates, thus enhancing patient care
during follow-up processes.
V. FUTURE WORK AND SUGGESTION
Deep learning, especially Convolutional
Neural Networks (CNNs), has been pivotal in analyzing retinal images. This is
crucial for the early detection of diseases like diabetic retinopathy,
glaucoma, and age-related macular degeneration (AMD). These networks excel in
extracting intricate features from medical images. This capability
significantly boosts diagnostic precision, surpassing traditional methods.
Expanding the dataset to encompass a
broader spectrum of eye diseases and demographic groups is essential. This move
enhances model resilience. Including diverse populations helps mitigate biases
inherent in current datasets. Exploring more advanced architectures, such as
Vision Transformers (ViTs) or hybrid models, could further refine detection
abilities. Employing specialized image processing techniques for specific eye
conditions also improves feature extraction. This, in turn, leads to enhanced
classification outcomes.
VI. CONCLUSION
Deep
learning has revolutionized the field of eye disease detection and classification,
marking a significant leap in diagnostic precision and speed. The application
of sophisticated algorithms, notably Convolutional Neural Networks (CNNs) and
transfer learning, has dramatically boosted early detection rates. This is
particularly evident in conditions such as diabetic retinopathy, cataracts, and
glaucoma.
VII. REFERENCES
[1] P. Ginsburg, “Machine learning
algorithms for identifying eye diseases using OCT imaging data,” J. Biomed.
Opt., vol. 11, no. 4, p. 041109, 2006.
[2] M. Farooq and A. Sattar, “Automated
detection of diabetic retinopathy and optical disc localization using image
classification techniques,” Comput. Biol. Med., vol. 66, pp. 47–57,
2015.
[3] D. S. Wei Ting et al.,
“Development and validation of a deep learning system for diabetic retinopathy
and related eye diseases using retinal images from multiethnic populations with
diabetes,” JAMA, vol. 318, no. 22, pp. 2211–2223, 2017.
[4] P. M. Burlina, N. Joshi, M. Pekala,
K. D. Pacheco, D. E. Freund, and N. M. Bressler, “Automated deep learning
diagnosis of age-related macular degeneration using optical coherence
tomography,” JAMA Ophthalmol., vol. 135, no. 11, pp. 1170–1176, 2017.
[5] S. Moccia, E. De Momi, S. El Hadji,
and L. S. Mattos, “Blood vessel segmentation algorithms—Review of methods,
datasets and evaluation metrics,” Comput. Methods Programs Biomed., vol.
158, pp. 71–91, 2018.
[6] F. Grassmann et al., “A deep
learning algorithm for prediction of age-related eye disease from color fundus
photography,” Sci. Rep., vol. 8, pp. 1–9, 2018.
[7] D. K. Prasad, D. Selvathi, and L.
Balasubramanian, “Deep neural network-based classification of early diabetic
retinopathy and glaucoma from retinal fundus images,” Comput. Biol. Med.,
vol. 111, p. 103351, 2019.
[8] X. Chen, Y. Xu, and X. Zhang,
“Glaucoma detection using deep convolutional neural networks,” in Med. Image
Comput. Comput.-Assist. Interv. – MICCAI 2015, pp. 795–802, 2015.
[9] Krishna, R. V. Babu, and V.
Sundararajan, “Simultaneous detection of diabetic retinopathy and glaucoma
using deep neural networks,” Pattern Recognit. Lett., vol. 125, pp.
78–85, 2019.
[10] M. Aun and M. Nazir, “Diabetic eye
disease localization using fast R-CNN and fuzzy k-means segmentation,” Procedia
Comput. Sci., vol. 170, pp. 248–255, 2020.
[11] A. Sarkari, H. Aghajan, and M.
Eslami, “Survey on deep learning-based diabetic eye disease detection using
fundus images,” Comput. Biol. Med., vol. 124, p. 103898, 2020.
[12] A. Chelaramani, S. Rao, and Y. R.
Reddy, “Multi-task learning for comprehensive eye disease analysis using fundus
images,” IEEE Access, vol. 8, pp. 185832–185844, 2020