Professional Documents
Culture Documents
Batch 18 Project - Report....
Batch 18 Project - Report....
Batch 18 Project - Report....
BELAGAVI-590018.
For the requirement of 7th Semester B.E in Computer Science & Engineering
Submitted by
Kruthika A (1KT20CS032)
Nagaveni N (1KT20CS048)
Piyush Kumar (1KT20CS053)
Shruthi BR(1KT20CS079)
CERTIFICATE
Diabetic Retinopathy (DR) is a prevalent complication of diabetes that affects the eyes, leading to
potential vision impairment or blindness if not diagnosed and treated in its early stages. It focuses
on developing an advanced classification system for DR disease levels by leveraging the power of
Graph Neural Networks (GNNs) to extract topological features from retinal images. Diabetic
retinopathy happens due to damage in blood vessels and is the prominent reason for blindness
worldwide. Clinical experts observe the fundus images to diagnose the disease, but it is often an
error prone and tedious task.
Computer-assisted techniques will help clinicians to detect the disease severity levels. In medical
imaging, experiments of automated diagnosis using CNN produce impressive results. Even though
disease classification tasks in retinal images via CNN face difficulty in retaining high quality
information at the output. A new deep learning methodology is proposed based on a graph
convolutional neural network (GCNN) .The proposed model aims to extract the essential retinal
image features effectively. The work focuses on extracting the features using a Variational
autoencoder and identifying the underlying topological correlations using GCNN.
The experiments are carried out using two datasets: Kaggle and EyePACS datasets. The
performance of the proposed model is evaluated using accuracy, U-kappa, sensitivity and
specificity metrics. The results outperform when compared with other state-of-the-art techniques.
Feature extraction is performed to capture essential information from the images, and these features
are used to construct a graph that represents the spatial relationships between different region of the
retina. .
I
ACKNOWLEDGEMENT
Kruthika A(1KT20CS032)
Nagaveni N(1KT20CS048)
Piyush Kumar(1KT20CS053)
Shruthi B R(1KT20CS079)
II
TABLE OF CONTENTS
ABSTRACT ........................................................... I
ACKNOWLEDGEMENT ...................................... II
Table of Contents
REFERENCES 19
LIST OF FIGURES
III
Classification of Diabetic Retinopathy Disease Level Using Graph Neural Network
CHAPTER 1
INTRODUCTION
For diabetic patients worldwide, diabetic retinopathy (DR) ends in premature blindness. Retinal disorders
should be diagnosed and treated using observations of blood vessels. Early diagnosis, fundus screening,
and timely intervention can prevent diabetic retinopathy, a severe cause of vision loss that may result in
blindness More precise early screening techniques are required in high-risk groups to reduce the threat of
vision loss by retinal disorders. Fundus pictures with expert interpretation are an acceptable screening
method for preventing blindness. The primary cause of diabetes’s clinical signs is an increase in blood
glucose levels, which can harm the retina’s blood vessels when present for an extended period. DR causes
no visual problems in the early stages. Complications in vision, such as floating patches or dark lines,
fuzziness or fluctuation, weak or dead areas of vision, and progressive blindness, may also develop as the
condition proceeds. Ophthalmologists must put in a lot of time and effort to manually detect diabetic
maculopathy. To improve the retinal image structure analysis, image processing methods are needed to
design Computer Assisted Diagnostic (CAD) systems. The introduction introduces the proposed deep
learning methodology based on a graph convolutional neural network (GCNN) for extracting topological
features from retinal images. It emphasizes the focus on extracting features using a Variational autoencoder
(VAE) and identifying underlying topological correlations using GCNN as key components of the
methodology.
Although deep learning effectively captures underlying patterns in data, there are many applications in
which data is represented graphically. Existing methods lack a unified objective for inter- and intra-
modality consistency learning and struggle with out-of-sample data. To address these challenges, a Graph
Embedding Contrastive Multi-modal Clustering network (GECMC) that integrates representation learning
and multi-modal clustering is proposed to enhance both capacities simultaneously . To address the
challenges of multi-view learning, the idea of Learnable Graph Convolutional Network and Feature Fusion
(LGCN-FF) that incorporates a learnable Graph Convolutional Network (GCN) and a feature fusion
network is proposed . The network combines features from various views using multiple sparse
autoencoders and a fully-connected network, resulting in a comprehensive representation that captures the
characteristics of all views. To tackle the limited interpretability and research gaps in multi-view learning
of Graph Convolutional Networks, novel framework called Interpretable Multi-view Graph Convolutional
Network (IMvGCN), which focuses on providing interpretability while solving multi-view learning
problems, is proposed .
To our knowledge, only a few kinds of research have been performed on retinal image classification using
graph neural networks. The images can be modeled as a graph structure where each pixel is represented as
a node . Deep 1 graph correlation network (DGCN) that utilizes a convolutional graph network can capture
natural correlations between independently learnt retinal image features . Graph convolutional networks
used for multi-label classification can be used to learn the complicated topology between lesion labels . The
vessel graph network (VGN) proposed by using a graph neural network (GNN) to transfer information along
vessel structures, have extracted hierarchical patterns in an image. However, works of GNN models that
exploit the global structure of vessel shape and their local appearances are limited.
PROBLEM STATEMENT
The problem statement of this paper is the detection and classification of diabetic retinopathy disease levels
using automated computer-assisted techniques. The current method of diagnosis by clinical experts through
observation of fundus images is often error-prone and tedious. Therefore, the proposed model aims to
extract essential retinal image features effectively using a graph convolutional neural network (GCNN) to
aid in the diagnosis of disease severity levels. The study focuses on extracting features using a Variational
autoencoder and identifying the underlying topological correlations using GCNN. The performance of the
proposed model is evaluated using accuracy, U-kappa, sensitivity, and specificity metrics.
1. Manual Diagnosis Limitations: Overcoming the error-prone and laborious nature of manual observation
by clinical experts for diabetic retinopathy disease level diagnosis.
2. Information Retention: Addressing the difficulty in preserving high-quality information at the output
in disease classification tasks using traditional CNN approaches.
3. Feature Extraction: Effectively extracting essential retinal image features to facilitate accurate diagnosis
Dept. of Computer science engineering 2023-24 Page No :5
Classification of Diabetic Retinopathy Disease Level Using Graph Neural Network
5. Performance Evaluation: Evaluating the model's performance using accuracy, U-kappa, sensitivity, and
specificity metrics to demonstrate its potential to outperform other state-of-the-art techniques.
1.1 OBJECTIVES
1.Proposing a novel methodology for classifying the severity levels of Diabetic Retinopathy Disease using
Graph Neural Networks (GNNs).
2. Experimenting with high-resolution retinal images from two datasets, including the Kaggle dataset and
the EyePACS dataset, to evaluate the performance of the proposed model.
3. Addressing the imbalance issue in the datasets through data augmentation techniques, such as horizontal
flip, width shift, height shift, fill mode, and zoom range, to increase the count of images.
4. Evaluating the performance of the proposed model in terms of sensitivity, specificity, accuracy, and U-
kappa scores, and comparing it with existing deep learning and machine learning-based models for retinal
disease classification.
5. Demonstrating the potential of the proposed Graph Neural Network model in capturing underlying
patterns in retinal images and identifying topological features for disease classification.
6. Providing a unified objective for inter- and intra-modality consistency learning and addressing challenges
in out-of-sample data through the proposed Graph Embedding Contrastive Multi-modal Clustering network
(GECMC)
1. Development of a deep learning methodology for automated diagnosis and classification of diabetic
retinopathy disease levels.
2. Utilization of graph neural networks to extract essential retinal image features and identify underlying
topological correlations.
3. Evaluation of the proposed model's performance using accuracy, U-kappa, sensitivity, and specificity
metrics.
5. Experimentation with two datasets: Kaggle and EyePACS, to validate the effectiveness of the proposed
methodology.
6. Addressing the limitations of manual observation by clinical experts in diagnosing diabetic retinopathy.
7. Overcoming the challenge of retaining high-quality information at the output in disease classification
tasks.
9. Enhancement of the understanding of diabetic retinopathy and its severity levels through the application
of advanced deep learning techniques to medical imaging data.
10. Provision of a framework for future research and development of computer-assisted techniques for the
diagnosis and management of retinal disorders.
1.3 MOTIVATION
The motivation behind this project is to address the critical need for improved diagnosis and classification
of diabetic retinopathy, a leading cause of blindness worldwide. Manual diagnosis is error-prone and
laborious, and there is a need for advanced deep learning methodologies to improve diagnostic accuracy.
By leveraging graph neural networks, the project aims to develop a computer-assisted technique that can
enhance diagnostic accuracy, enable earlier interventions, and ultimately improve patient outcomes. The
potential for automation and the advancement of medical imaging techniques further drive the motivation
for this project. Improved diagnostic accuracy can enable earlier interventions and ultimately improve
patient outcomes. The project aims to contribute to the broader field of healthcare technology and potentially
pave the way for similar advancements in other areas of healthcare. Overall, the motivation for this project
stems from the potential to significantly impact the field of diabetic retinopathy diagnosis and ultimately
improve patient care and outcomes.
CHAPTER 2
LITERATURE SURVEY
[1] Paper Title: "Detection of diabetic retinopathy using convolution neural network"
Author: Hare Shyam Sharama , Ajith Singh Chandel
Publication Year: 2019
Methodology: Convolutional neural network (CNNs) are widely used in pattern- and image-
recognition problems as they have a number of advantages compared to other techniques. Though it
require huge computation power to train a CNN but while testing the results are generated very fast.
The model was trained on a diverse dataset of blood cell images .
Remarks: A model is presented for classification of DR stages based on the severity using colour fundus images
.The performance of the model is assessed using different metrics.
Remarks: the paper provides a comprehensive review of the methods used in the detection of
diabetic retinopathy using CAD systems.
[3] Paper Title: " Deep Learning Techniques for Diabetic Retinopathy Classification "
Methodology: The literature survey paper provides an overview of the latest methods used in diabetic
retinopathy classification and detection. It covers classical techniques such as Adaboost, Random
Forest, and SVM, as well as deep learning methods. The paper highlights the limitations of classical
techniques and the importance of improving deep learning models.
Remarks: It also emphasizes the use of data augmentation to reduce overfitting in model training.
Overall, the survey aims to provide insights into the current state of research in this field
[4] Paper Title: " Dynamic Graph Clustering Learning for Unsupervised Diabetic Retinopathy Classificatio
Methodology: . The proposed Dynamic Graph Clustering Learning (DGCL) method offers a novel
approach to unsupervised classification of diabetic retinopathy, leveraging multi-structural feature fusion
and consistency smoothing clustering modules.
Remarks: the research contributes to advancing the field of medical image classification and holds
promise for future applications in clinical settings .
[5] Paper Title :” Classification of Diabetic Retinopathy Severity Based on GCA Attention Mechanism”
[6] Paper Title : “A Deep Learning Ensemble Approach for Diabetic Retinopathy Detection”
Remarks: The experimental results show that the proposed model detects all the stages of DR unlike
the current methods and performs better compared to state-of-the-art methods on the same Kaggle
dataset.
[7] Paper Title: " Deep Graph Correlation Network on Retinal Images Without Manual Annotations"
Author: Guanghua Zhang, Bin Sun
Methodology: This study aims to economize manual power and proposes a deep graph correlation
network (DGCN) to develop automated diabetic retinopathy grading without any professional
annotations. DGCN involves the novel deep learning algorithm of a graph convolutional network to
exploit inherent correlations from independent retinal image features learned by a convolutional neural
network.
Remarks: The grading capability of this study is close to that of retina specialists, but superior to that
of trained graders, which demonstrates that the proposed DGCN provides an innovative route for
automated diabetic retinopathy grading and other computer-aided diagnostic systems.
[8] Paper Title: “Detecting and staging diabetic retinopathy in retinal images using multi-branch CNN”
Methodology : The authors propose a CNN-based approach, DeepRoot, to detect and grade diabetic
retinopathy in retinal images. The proposed architecture includes a main branch and two side branches
to capture fine-level features. The model is trained, validated, and tested on retinal image datasets,
including self-collected samples from a hospital.
Remarks : This project introduces a novel CNN architecture, DeepRoot, for accurate detection and
grading of diabetic retinopathy in retinal images, demonstrating high sensitivity and promising
performance. The methodology involves training, validation, and testing on diverse retinal image
datasets, including real-world samples from a hospital, showcasing the model's robustness.
[9] Paper Title: “Diabetic Retinopathy Detection”
Remarks : As we used Convolutional neural network methodology, every input image will be
classified with the highest accuracy assuring no misclassification happens. We first trained the
CNN to recognize lines, edges, corners etc. from the retinal image followed by recognizing small parts
of a single feature which is then followed by recognizing a whole feature.
[10] Paper Title: “Diabetic retinopathy detection through deep learning techniques”
Remarks : Automated screening systems significantly reduce the time required to determine diagnoses,
saving effort and costs for ophthalmologists and result in the timely treatment of patients. Automated
systems for DR detection play an important role in detecting DR at an early stage.
CHAPTER 3
REQUIREMENTS SPECIFICATION
1. Data Collection and Preprocessing: Gathering retinal images and preprocessing them to ensure
uniformity and quality for analysis .
2. Feature Extraction: Implementing methods to extract essential retinal image features using graph neural
networks and variational autoencoders.
3. Disease Severity Classification: Developing algorithms to classify diabetic retinopathy severity levels
based on extracted features .
4. Model Training and Validation: Training the graph neural network model using labeled data and
validating its performance using appropriate metrics such as accuracy, sensitivity, and specificity .
5. Integration with Clinical Workflow: Designing the system to seamlessly integrate with clinical
workflows, allowing for efficient utilization by healthcare professionals.
6. User Interface: Creating a user-friendly interface for clinicians to interact with the system, input data,
and interpret results.
7. Scalability and Performance: Ensuring that the system can handle a large volume of retinal images
efficiently and provide timely results.
8. Security and Compliance: Implementing measures to ensure data security, privacy, and compliance
with healthcare regulations and standards.
10. Maintenance and Support: Establishing protocols for ongoing maintenance, updates, and user support
to ensure the system's continued effectiveness and relevance.
1. Performance: The system should demonstrate high performance, with efficient processing of retinal
images and quick classification of diabetic retinopathy severity levels to support timely interventions and
decision-making .
2. Reliability: The system must be reliable, providing consistent and accurate results in diagnosing diabetic
retinopathy severity levels to ensure patient safety and trust in the system .
3. Usability: The user interface should be intuitive and user-friendly, allowing healthcare professionals to
interact with the system easily and interpret the results without extensive training .
4. Security: The system must adhere to stringent security measures to protect patient data, ensuring
compliance with healthcare privacy regulations and safeguarding sensitive medical information .
5. Scalability: The system should be designed to scale effectively, accommodating a growing number of
retinal images and users while maintaining performance and reliability .
6. Interoperability: The system must be interoperable with existing healthcare systems and standards,
facilitating seamless integration into clinical workflows and data exchange with other medical applications.
CHAPTER 4
METHODOLOGY
1. Data Collection and Preprocessing: The project likely involves the collection of retinal images and
preprocessing them to ensure uniformity and quality for analysis .
2. Feature Extraction: The methodology includes the extraction of essential retinal image features using
a variational autoencoder and the identification of underlying topological correlations using graph neural
networks.
3. Model Development: A new deep learning methodology based on a graph convolutional neural network
(GCNN) is proposed to effectively extract essential retinal image features .
4. Dataset Utilization: The experiments are carried out using two datasets: Kaggle and EyePACS datasets,
indicating the utilization of real-world data for model training and evaluation .
5. Performance Evaluation: The performance of the proposed model is evaluated using metrics such as
accuracy, U-kappa, sensitivity, and specificity, and the results outperform other state-of-the-art techniques
.
6. Experimental Observations: The experiments involve training the model on labeled data, utilizing the
trained model to label unlabeled data, and testing the model's performance on the unlabeled data using
techniques such as early stopping and multiple epochs .
7. Clinical Application: The proposed methodology aims to assist clinicians in classifying retinal images
according to their diabetic retinopathy severity based on their abnormalities, addressing the need for
computer-assisted diagnosis in early analysis of diabetic retinopathy.
ARCHITECTURE DIAGRAM
1.Input Image:
Description: The system begins with input images capturing regions of interest related to symptoms
of blindness.
Components:
1. Image Preprocessing:
Description: Raw images undergo preprocessing to enhance quality and standardize input for
subsequent stages.
Components:
Description : Segmentation isolates specific regions within the images related to symptoms of
blindness.
Components: Techniques such as semantic segmentation to identify and separate relevant features
Components: neural network with three primary color components of the image, namely red, green,
and blue as inputs, is used to identify and segment retinal blood vessels.
4. Microaneurysm segmentation:
Components:
5. Classification:
Description: The final stage involves classifying the input images based on learned features to
Components:
7. Result Output:
Description: The system generates output indicating the detected its severity, and possibly
recommended actions.
Components:
REFERENCES
[1] S. Sundar and S. Sumathy, ‘‘An effective deep learning model for grading abnormalities in retinal fundus
images using variational auto-encoders,’’ Int. J. Imag. Syst. Technol., vol. 33, no. 1, pp. 92–107, Jan. 2023.
[2] S. Kumari, P. Venkatesh, N. Tandon, R. Chawla, B. Takkar, and A. Kumar, ‘‘Selfie fundus imaging for
diabetic retinopathy screening,’’ Eye, vol. 36, no. 10, pp. 1988–1993, Oct. 2022.
[3] A. K. Gangwar and V. Ravi, ‘‘Diabetic retinopathy detection using transfer learning and deep learning,’’
in Evolution in Computational Intelligence: Frontiers in Intelligent Computing: Theory and Applications,
vol. 1. Singapore: Springer, 2021.
[4] R. Gargeya and T. Leng, ‘‘Automated identification of diabetic retinopathy using deep learning,’’
Ophthalmology, vol. 124, no. 7, pp. 962–969, Jul. 2017.
[5] X. Li, X. Hu, L. Yu, L. Zhu, C. Fu, and P. Heng, ‘‘CANet: Cross-disease attention network for joint
diabetic retinopathy and diabetic macular edema grading,’’ IEEE Trans. Med. Imag., vol. 39, no. 5, pp. 1483–
1493, May 2020.
[6] R. Pires, S. Avila, J. Wainer, E. Valle, M. D. Abramoff, and A. Rocha, ‘‘A data-driven approach to
referable diabetic retinopathy detection,’’ Artif. Intell. Med., vol. 96, pp. 93–106, May 2019.
[7] J. Y. Choi, T. K. Yoo, J. G. Seo, J. Kwak, T. T. Um, and T. H. Rim, ‘‘Multicategorical deep learning
neural network to classify retinal images: A pilot study employing small database,’’ PLoS ONE, vol. 12, no.
11, Nov. 2017, Art. no. e0187336.
[8] S. Sumod and S. Sumathy, ‘‘Transfer learning approach in deep neural networks for uterine fibroid
detection,’’ Int. J. Comput. Sci. Eng., vol. 25, pp. 52–63, Jan. 2022.
[9] W. Xia, T. Wang, Q. Gao, M. Yang, and X. Gao, ‘‘Graph embedding contrastive multi-modal
representation learning for clustering,’’ IEEE Trans. Image Process., vol. 32, pp. 1170–1183, 2023.
[10] Z. Chen, L. Fu, J. Yao, W. Guo, C. Plant, and S. Wang, ‘‘Learnable graph convolutional network and
feature fusion for multi-view learning,’’ Inf. Fusion, vol. 95, pp. 109–119, Jul. 2023. 15