You are on page 1of 9

Automated Diabetic

Retinopathy Detection
Using Deep Learning
Karuna Mudagal 2KA20CS018

Prajwal N Bhasme 2KA20CS037

Suhail M Mullanavar 2KA20CS055

Vidya V Hampiholi 2KA20CS063


Introduction
Diabetes is a disease that has a major prevalence in our population. One of
the complications resulting from the disease is diabetic retinopathy (DR).

A commonly used method of diagnosing DR is by analyzing fundus


images of the eye and classifying the condition, if present, into one of the
following categories: Mild, Moderate or severe non-proliferative diabetic
retinopathy and proliferative diabetic retinopathy.

The aim of this study is to use this technology along with a library of
classified DR images to first, train a CNN to recognize diabetic and non-
diabetic images and secondly, visualize the trained CNN, to search for any
new indicators of DR that may not have been used by physicians before to
diagnose the disease. These factors may help in increasing the accuracy of
diagnose and allowing detection at an earlier stage.
Literature Survey
​ L.
S
Reference Technology Used Limitations​
NO

Convolutional Neural
John et al (2022)., Title of Paper, Limited dataset size,
1 Networks (CNN)
lack of diversity.
Journal of AI in Medicine. Image Processing.

Smith and Brown (2022).,


Support Vector Reliance on
2​ S. Smith, P. Brown, Another Paper, Machines (SVM). handcrafted features.
Conference on ML.
Yang and Kim et al (2022).,
Deep Learning (e.g., Challenges in
3​ Y. Yang, K. Kim, Yet Another Paper, LSTM). interpretability.
IEEE Transactions on Biomed.
Johnson, Patel et al (2022)., Transfer Learning,
Deep Learning,
Recent Advances, International Limited explanation of
4 Optical
model decisions.
Journal of computer Vision. Coherence
Tomography.​
Problem Statement

 The problem statement revolves around the need to improve screening and diagnostic methods, increase
access to eye care services, and raise awareness and education regarding regular eye exams for individuals
with diabetes. Diabetic retinopathy is a common complication of diabetes, affecting the eyes and
potentially leading to blindness if not managed effectively. Many individuals with diabetes, particularly
in undeserved populations, lack access to timely and effective screening, diagnosis, and treatment for
diabetic retinopathy.
Objectives
01 02
To identify and decide Diabetic Retinopathy (DR) using Identifying the severity of Diabetic Retinopathy (DR).
digital fundus images.

03
To determine the actual positions and dimensions of the
spots.
Methodology

1. Data Collection:
Retinal images, typically obtained through fundus
photography, are collected. These images serve as
the dataset for training and testing the deep
learning model.

4. Training the Model: 6. Deployment:


2. Data Preprocessing:
The model is trained using a labeled dataset, where Once the model demonstrates satisfactory
Raw retinal images are preprocessed to enhance
each retinal image is annotated with information performance, it can be deployed for automated
quality and remove artifacts. This may include
about the presence and severity of diabetic diabetic retinopathy screening. This involves
resizing, normalization, and other techniques to
retinopathy. The model learns to recognize patterns integrating the model into a system that can
standardize the data.
and features associated with different stages of the process new retinal images and provide diagnostic

3. Deep Learning Model Architecture: disease. predictions.

5. Validation and Fine-Tuning: 7. Post-Processing and Reporting:


A deep neural network, often a convolutional
The trained model is validated on a separate dataset Post-processing steps may be applied to refine the
neural network (CNN), is designed to process and
to ensure generalizability. Fine-tuning may be model's predictions. The final output includes a
analyze retinal images. CNNs are particularly
performed to optimize the model's performance based diagnostic report indicating the likelihood and
effective for image-related tasks due to their ability
on validation results. severity of diabetic retinopathy.
to learn hierarchical features.
Requirements

S O F T WA R E H A R D WA R E

 Deep Learning Frameworks.  High-Performance GPUs.

 CUDA and cuDNN.  Memory(RAM).

 Python and Libraries.  Storage(SSD).

 Development Environment.

 Version Control.

 Containerization.
Expected Outcomes

 Early Detection of Diabetic retinopathy.

 Improved Accuracy of Diabetic retinopathy.

 Reduced Workload for Healthcare Providers.

 Prevention of Vision Loss.


Thank you

You might also like