Sunteți pe pagina 1din 17

Diabetic Retinopathy(DR) using Convolutional

Neural Network(CNN)

Guided By:
Prof. Rina Bora

Presented By:
Sayali Sharad Manjrekar – 20
Ashish Dnyaneshwar Parulekar – 29
Malay Hitendra Patel - 30
Contents
• Introduction
• Literature Survey
• Problem Statement
• Proposed methodology
• Software Details
• Applications
• Conclusion
• References
Introduction
1. Diabetic Retinopathy(DR) is when a person suffering from diabetes from
more than 10 years.
2. Due to this, blood vessels in retina gets damaged, and suffering from a
blurred vision, difficulty in seeing colours or in some cases even total loss
of vision.
3. Retinal surgeries can relieve the symptoms, but efficient way to tackle
Diabetic Retinopathy(DR) is controlling diabetes and managing early
symptoms of Diabetic Retinopathy(DR), which can be done by having
eyes checked atleast once annually. Diabetic Retinopathy(DR) is a
leading cause of blindness.
4. The manual process for studying the fundus images, is a time consuming
approach and needs expertise to do it. Thus to minimize the time and to
detect it more accurately, the proposed system uses deep Convolutional
Neural Network(CNN) architecture.
Introduction
5. CNN is a Convolutional Neural Network which learns to create filters on
the input image, which can then effectively classify the images based on
what filters it passes activating those specific neurons inside each layer.
Earlier layers generally create filters which are able to detect edges and
geometric shapes while the later deeper layers have filters which identify
more details like faces, eyes etc.This technique is very effective in image
recognition and classification.

6. Proposed system is used to classify diabetic retinopathy into 5 stages,


including: 1) Stage I: No apparent retinopathy, 2) Stage II: Mild None-
Proliferative Diabetic Retinopathy (NPDR), 3) Stage III: Moderate
NPDR, 4) Stage IV: Severe NPDR, and 5) Stage V: Proliferative diabetic
retinopathy.
Literature Survey
1. Xiaoliang Wang, (2018) decribed about the Diabetic Retinopathy(DR)
stage classification, and used deep learning based Convolutional Neural
Network and deployed VGG16, AlexNet and Inception V3 architectures
for DR stage classification. In their study, 3 Convolutional Network
architecture as mentioned above are trained to produce predictive models
using Cross Validation Process, their efforts will provide a useful
software-based tools for opthalmologists to evaluate the severity level of
diabetes mellitus, by recognizing different DR stage.

Training Algorithm Avg. CV Accuracy

Alexnet 37.43%
VGG16 50.03%
InceptionNet V3 63.23%
Literature Survey
2. Darshit Doshi, (2016) described about automatic diagonsis of Diabetic
Retinopathy(DR) into its different stages using Deep Learning, and
implementaion of GPU accelerated deep Convolutional Neural Network
to automatically Diagonose and classify high resolutional retinal images
into 5 stages of the disease based on severity. Data preprocessing
techniques were performed using Image magic(command line tool for
image processing) and the python library OpenCV and results were
evaluated using a quadratic weighted kappa metric.
Problem Statement
Given an image of an eye of the patient, to create an automated detection
system for the disease which can benefit the population, by early and easy
detection of disease, and classify in following classes as 1) Stage I: No
apparent retinopathy, 2) Stage II: Mild None-Proliferative Diabetic
Retinopathy (NPDR), 3) Stage III: Moderate NPDR, 4) Stage IV: Severe
NPDR, and 5) Stage V: Proliferative diabetic retinopathy.
Proposed methodology
• CNN extends the regular neural
• networks by adding the operations of convolution, nonlinearity and sub-sampling. The purpose of
convolution is to
• extract features from the input images. By convoluting the
• input images with some specially chosen small square
• matrices, certain image processing effects, such as edge
• detection, sharpening and blurring could be realized. Another
• operation called Rectified Linear Unit (ReLU) that could be
• used after every convolution operation is a non-linear
• operation by replacing all negative pixel values in the feature
• map by zero. The purpose of ReLU is to introduce
• nonlinearity. The third operation called pooling or subsampling reduces the dimensionality of each
feature map
• while retaining the most important information. It is realized
• by taking and storing for example, the max, average or sum of
• a sub-region in the feature map. After adding an appropriate
• number of layers of these three operations, the output feature
• map will be connected to a classical neural network to
• complete the classification task.
• Our Solution consisted of two main steps namely Image preprocessing and Retraining the the
Inception V3.
• A. Image Preprocessing
• Image preprocessing can be divided into three subtasks-
• 1. Down sampling the images to a specific radius.
• 2. Subtracting Local Average Color from the images.
• 3. Cropping the image borders.

• Due to the extremely high variability in the size of images, all the images were downsized to a
common radius of 200 pixels and 500 pixels. Images with the same radii were used retrain the
Inception network at a time.
• After down sampling of the images, the local average color was subtracted from the complete
image to highlight the essential features of the image such as lesions, Microaneurysms, swelling of
the blood vessels etc. The subtraction of the local average color was also necessary because images
were captured at different color spectrums and the background color must not be used by the
classifier as a feature to identify the severity of DR.
• The last step of the Image preprocessing was to clip the image to remove its boundary effects.
PROPOSED METHODLOGY
• B. Retraining the InceptionV3
• Transfer learning means to use a model that has already
been trained on other images. But now we have different
set of images that we want our classifier to give prediction
on, then we train only the last layer of the architecture. The
basic logic behind transfer learning is that each layer on the
pre-trained InceptionV3 network [12] determines specific
shapes particular to the worldly objects such as edges,
circles etc.
• We use this information for prediction on our new classes.
We load the pre-trained model and remove the old final
layer and train a new one on the eye images. [8][11]. We
select equal images from each class so that there is no class
imbalance.
PROPOSED METHODOLOGY
• GoogLeNet utilized channel concatenation: concatenating the
• channels obtained from 1 × 1 convolution, 3 × 3 convolution,
• 5 × 5 convolution and pooling together while keeping the
• height and width of each channel unchanged. To avoid the
• exponentially increase of the mathematical operations that
• have to be done for channel concatenation, 1 × 1 “bottleneck”
• layers that have shallower channel depths are applied before
• applying 3 × 3 or 5 × 5 convolution. The introduction of these
• “bottleneck” layers reduces the number of mathematical
• operations for a particular convolution by a factor of 10. As a
• result of channel concatenation and 1 × 1 convolution
• “bottleneck” layers, GoogleNet employed only 5 million
• parameters, which represented a 12× reduction with respect to
• its predecessor AlexNet [14], which used 60 million
• parameters. Furthermore, VGGNet employed about 3× more
• parameters than AlexNet. Compared to the GoogleNet [17],
• InceptionNet V3 [18] utilizes factorizing convolutions
• strategies to further increase computational efficiency. Figure
• 4 exemplifies the inception module of InceptionNet V3
• leverages spatial factorization into asymmetric convolutions
• to save computation cost further.
Software Details
• What is Google Colab?
Google Colab is a free cloud service and now it supports free GPU! You
can:
• improve your Python programming language coding skills.
• develop deep learning applications using popular libraries such as Keras,
TensorFlow, PyTorch, and OpenCV.
• Since Colab is working on your own Google Drive,
Applications
1. Computer vision
2. Scene labelling
3. Image Classification
4. Action Recognition
5. Human Pose Estimation
6. Document Analysis
7. Feature User Interface for Healthcare Professional i.e. Opthalmologists
Conclusion
References
• - Diabetic Retinopathy stage classification
using Convolutional Neural Network, BY
Xiaoliang Wang,Yongjin Lu, Yujuan Wang, Wei-
Bang Chen.
• - Diabetic Retinopathy Detection using Deep
Convolutional Neural Networks, BY Darshit
Doshi, Aniket Shenoy, Deep Sidhpura, Dr.
Prachi Gharpure.

S-ar putea să vă placă și